TCLR: Temporal contrastive learning for video representation
Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct...
Ausführliche Beschreibung
Gespeichert in:
| Autor*in: |
Dave, Ishan [verfasserIn] |
|---|
| Format: |
E-Artikel |
|---|---|
| Sprache: |
Englisch |
| Erschienen: |
2022transfer abstract |
|---|
| Schlagwörter: |
|---|
| Übergeordnetes Werk: |
Enthalten in: Editorial Board - 2016, CVIU, San Diego, Calif |
|---|---|
| Übergeordnetes Werk: |
volume:219 ; year:2022 ; pages:0 |
| Links: |
|---|
| DOI / URN: |
10.1016/j.cviu.2022.103406 |
|---|
| Katalog-ID: |
ELV057712875 |
|---|
| LEADER | 01000caa a22002652 4500 | ||
|---|---|---|---|
| 001 | ELV057712875 | ||
| 003 | DE-627 | ||
| 005 | 20230626045616.0 | ||
| 007 | cr uuu---uuuuu | ||
| 008 | 220808s2022 xx |||||o 00| ||eng c | ||
| 024 | 7 | |a 10.1016/j.cviu.2022.103406 |2 doi | |
| 028 | 5 | 2 | |a /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica |
| 035 | |a (DE-627)ELV057712875 | ||
| 035 | |a (ELSEVIER)S1077-3142(22)00037-6 | ||
| 040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
| 041 | |a eng | ||
| 082 | 0 | 4 | |a 630 |q VZ |
| 082 | 0 | 4 | |a 640 |q VZ |
| 082 | 0 | 4 | |a 610 |q VZ |
| 100 | 1 | |a Dave, Ishan |e verfasserin |4 aut | |
| 245 | 1 | 0 | |a TCLR: Temporal contrastive learning for video representation |
| 264 | 1 | |c 2022transfer abstract | |
| 336 | |a nicht spezifiziert |b zzz |2 rdacontent | ||
| 337 | |a nicht spezifiziert |b z |2 rdamedia | ||
| 338 | |a nicht spezifiziert |b zu |2 rdacarrier | ||
| 520 | |a Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. | ||
| 520 | |a Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. | ||
| 650 | 7 | |a 68T45 |2 Elsevier | |
| 650 | 7 | |a 68T07 |2 Elsevier | |
| 650 | 7 | |a 68T30 |2 Elsevier | |
| 700 | 1 | |a Gupta, Rohit |4 oth | |
| 700 | 1 | |a Rizve, Mamshad Nayeem |4 oth | |
| 700 | 1 | |a Shah, Mubarak |4 oth | |
| 773 | 0 | 8 | |i Enthalten in |n Elsevier |t Editorial Board |d 2016 |d CVIU |g San Diego, Calif |w (DE-627)ELV019616708 |
| 773 | 1 | 8 | |g volume:219 |g year:2022 |g pages:0 |
| 856 | 4 | 0 | |u https://doi.org/10.1016/j.cviu.2022.103406 |3 Volltext |
| 912 | |a GBV_USEFLAG_U | ||
| 912 | |a GBV_ELV | ||
| 912 | |a SYSFLAG_U | ||
| 912 | |a SSG-OLC-PHA | ||
| 951 | |a AR | ||
| 952 | |d 219 |j 2022 |h 0 | ||
| author_variant |
i d id |
|---|---|
| matchkey_str |
daveishanguptarohitrizvemamshadnayeemsha:2022----:creprlotatvlannfrie |
| hierarchy_sort_str |
2022transfer abstract |
| publishDate |
2022 |
| allfields |
10.1016/j.cviu.2022.103406 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica (DE-627)ELV057712875 (ELSEVIER)S1077-3142(22)00037-6 DE-627 ger DE-627 rakwb eng 630 VZ 640 VZ 610 VZ Dave, Ishan verfasserin aut TCLR: Temporal contrastive learning for video representation 2022transfer abstract nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. 68T45 Elsevier 68T07 Elsevier 68T30 Elsevier Gupta, Rohit oth Rizve, Mamshad Nayeem oth Shah, Mubarak oth Enthalten in Elsevier Editorial Board 2016 CVIU San Diego, Calif (DE-627)ELV019616708 volume:219 year:2022 pages:0 https://doi.org/10.1016/j.cviu.2022.103406 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U SSG-OLC-PHA AR 219 2022 0 |
| spelling |
10.1016/j.cviu.2022.103406 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica (DE-627)ELV057712875 (ELSEVIER)S1077-3142(22)00037-6 DE-627 ger DE-627 rakwb eng 630 VZ 640 VZ 610 VZ Dave, Ishan verfasserin aut TCLR: Temporal contrastive learning for video representation 2022transfer abstract nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. 68T45 Elsevier 68T07 Elsevier 68T30 Elsevier Gupta, Rohit oth Rizve, Mamshad Nayeem oth Shah, Mubarak oth Enthalten in Elsevier Editorial Board 2016 CVIU San Diego, Calif (DE-627)ELV019616708 volume:219 year:2022 pages:0 https://doi.org/10.1016/j.cviu.2022.103406 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U SSG-OLC-PHA AR 219 2022 0 |
| allfields_unstemmed |
10.1016/j.cviu.2022.103406 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica (DE-627)ELV057712875 (ELSEVIER)S1077-3142(22)00037-6 DE-627 ger DE-627 rakwb eng 630 VZ 640 VZ 610 VZ Dave, Ishan verfasserin aut TCLR: Temporal contrastive learning for video representation 2022transfer abstract nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. 68T45 Elsevier 68T07 Elsevier 68T30 Elsevier Gupta, Rohit oth Rizve, Mamshad Nayeem oth Shah, Mubarak oth Enthalten in Elsevier Editorial Board 2016 CVIU San Diego, Calif (DE-627)ELV019616708 volume:219 year:2022 pages:0 https://doi.org/10.1016/j.cviu.2022.103406 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U SSG-OLC-PHA AR 219 2022 0 |
| allfieldsGer |
10.1016/j.cviu.2022.103406 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica (DE-627)ELV057712875 (ELSEVIER)S1077-3142(22)00037-6 DE-627 ger DE-627 rakwb eng 630 VZ 640 VZ 610 VZ Dave, Ishan verfasserin aut TCLR: Temporal contrastive learning for video representation 2022transfer abstract nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. 68T45 Elsevier 68T07 Elsevier 68T30 Elsevier Gupta, Rohit oth Rizve, Mamshad Nayeem oth Shah, Mubarak oth Enthalten in Elsevier Editorial Board 2016 CVIU San Diego, Calif (DE-627)ELV019616708 volume:219 year:2022 pages:0 https://doi.org/10.1016/j.cviu.2022.103406 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U SSG-OLC-PHA AR 219 2022 0 |
| allfieldsSound |
10.1016/j.cviu.2022.103406 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica (DE-627)ELV057712875 (ELSEVIER)S1077-3142(22)00037-6 DE-627 ger DE-627 rakwb eng 630 VZ 640 VZ 610 VZ Dave, Ishan verfasserin aut TCLR: Temporal contrastive learning for video representation 2022transfer abstract nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. 68T45 Elsevier 68T07 Elsevier 68T30 Elsevier Gupta, Rohit oth Rizve, Mamshad Nayeem oth Shah, Mubarak oth Enthalten in Elsevier Editorial Board 2016 CVIU San Diego, Calif (DE-627)ELV019616708 volume:219 year:2022 pages:0 https://doi.org/10.1016/j.cviu.2022.103406 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U SSG-OLC-PHA AR 219 2022 0 |
| language |
English |
| source |
Enthalten in Editorial Board San Diego, Calif volume:219 year:2022 pages:0 |
| sourceStr |
Enthalten in Editorial Board San Diego, Calif volume:219 year:2022 pages:0 |
| format_phy_str_mv |
Article |
| institution |
findex.gbv.de |
| topic_facet |
68T45 68T07 68T30 |
| dewey-raw |
630 |
| isfreeaccess_bool |
false |
| container_title |
Editorial Board |
| authorswithroles_txt_mv |
Dave, Ishan @@aut@@ Gupta, Rohit @@oth@@ Rizve, Mamshad Nayeem @@oth@@ Shah, Mubarak @@oth@@ |
| publishDateDaySort_date |
2022-01-01T00:00:00Z |
| hierarchy_top_id |
ELV019616708 |
| dewey-sort |
3630 |
| id |
ELV057712875 |
| language_de |
englisch |
| fullrecord |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">ELV057712875</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230626045616.0</controlfield><controlfield tag="007">cr uuu---uuuuu</controlfield><controlfield tag="008">220808s2022 xx |||||o 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1016/j.cviu.2022.103406</subfield><subfield code="2">doi</subfield></datafield><datafield tag="028" ind1="5" ind2="2"><subfield code="a">/cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)ELV057712875</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(ELSEVIER)S1077-3142(22)00037-6</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">630</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">640</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">610</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Dave, Ishan</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">TCLR: Temporal contrastive learning for video representation</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2022transfer abstract</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zzz</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">z</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zu</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR.</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR.</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">68T45</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">68T07</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">68T30</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Gupta, Rohit</subfield><subfield code="4">oth</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Rizve, Mamshad Nayeem</subfield><subfield code="4">oth</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Shah, Mubarak</subfield><subfield code="4">oth</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="n">Elsevier</subfield><subfield code="t">Editorial Board</subfield><subfield code="d">2016</subfield><subfield code="d">CVIU</subfield><subfield code="g">San Diego, Calif</subfield><subfield code="w">(DE-627)ELV019616708</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:219</subfield><subfield code="g">year:2022</subfield><subfield code="g">pages:0</subfield></datafield><datafield tag="856" ind1="4" ind2="0"><subfield code="u">https://doi.org/10.1016/j.cviu.2022.103406</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ELV</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-PHA</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">219</subfield><subfield code="j">2022</subfield><subfield code="h">0</subfield></datafield></record></collection>
|
| author |
Dave, Ishan |
| spellingShingle |
Dave, Ishan ddc 630 ddc 640 ddc 610 Elsevier 68T45 Elsevier 68T07 Elsevier 68T30 TCLR: Temporal contrastive learning for video representation |
| authorStr |
Dave, Ishan |
| ppnlink_with_tag_str_mv |
@@773@@(DE-627)ELV019616708 |
| format |
electronic Article |
| dewey-ones |
630 - Agriculture & related technologies 640 - Home & family management 610 - Medicine & health |
| delete_txt_mv |
keep |
| author_role |
aut |
| collection |
elsevier |
| remote_str |
true |
| illustrated |
Not Illustrated |
| topic_title |
630 VZ 640 VZ 610 VZ TCLR: Temporal contrastive learning for video representation 68T45 Elsevier 68T07 Elsevier 68T30 Elsevier |
| topic |
ddc 630 ddc 640 ddc 610 Elsevier 68T45 Elsevier 68T07 Elsevier 68T30 |
| topic_unstemmed |
ddc 630 ddc 640 ddc 610 Elsevier 68T45 Elsevier 68T07 Elsevier 68T30 |
| topic_browse |
ddc 630 ddc 640 ddc 610 Elsevier 68T45 Elsevier 68T07 Elsevier 68T30 |
| format_facet |
Elektronische Aufsätze Aufsätze Elektronische Ressource |
| format_main_str_mv |
Text Zeitschrift/Artikel |
| carriertype_str_mv |
zu |
| author2_variant |
r g rg m n r mn mnr m s ms |
| hierarchy_parent_title |
Editorial Board |
| hierarchy_parent_id |
ELV019616708 |
| dewey-tens |
630 - Agriculture 640 - Home & family management 610 - Medicine & health |
| hierarchy_top_title |
Editorial Board |
| isfreeaccess_txt |
false |
| familylinks_str_mv |
(DE-627)ELV019616708 |
| title |
TCLR: Temporal contrastive learning for video representation |
| ctrlnum |
(DE-627)ELV057712875 (ELSEVIER)S1077-3142(22)00037-6 |
| title_full |
TCLR: Temporal contrastive learning for video representation |
| author_sort |
Dave, Ishan |
| journal |
Editorial Board |
| journalStr |
Editorial Board |
| lang_code |
eng |
| isOA_bool |
false |
| dewey-hundreds |
600 - Technology |
| recordtype |
marc |
| publishDateSort |
2022 |
| contenttype_str_mv |
zzz |
| container_start_page |
0 |
| author_browse |
Dave, Ishan |
| container_volume |
219 |
| class |
630 VZ 640 VZ 610 VZ |
| format_se |
Elektronische Aufsätze |
| author-letter |
Dave, Ishan |
| doi_str_mv |
10.1016/j.cviu.2022.103406 |
| dewey-full |
630 640 610 |
| title_sort |
tclr: temporal contrastive learning for video representation |
| title_auth |
TCLR: Temporal contrastive learning for video representation |
| abstract |
Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. |
| abstractGer |
Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. |
| abstract_unstemmed |
Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR. |
| collection_details |
GBV_USEFLAG_U GBV_ELV SYSFLAG_U SSG-OLC-PHA |
| title_short |
TCLR: Temporal contrastive learning for video representation |
| url |
https://doi.org/10.1016/j.cviu.2022.103406 |
| remote_bool |
true |
| author2 |
Gupta, Rohit Rizve, Mamshad Nayeem Shah, Mubarak |
| author2Str |
Gupta, Rohit Rizve, Mamshad Nayeem Shah, Mubarak |
| ppnlink |
ELV019616708 |
| mediatype_str_mv |
z |
| isOA_txt |
false |
| hochschulschrift_bool |
false |
| author2_role |
oth oth oth |
| doi_str |
10.1016/j.cviu.2022.103406 |
| up_date |
2024-07-06T16:56:34.058Z |
| _version_ |
1803849557933555712 |
| fullrecord_marcxml |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">ELV057712875</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230626045616.0</controlfield><controlfield tag="007">cr uuu---uuuuu</controlfield><controlfield tag="008">220808s2022 xx |||||o 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1016/j.cviu.2022.103406</subfield><subfield code="2">doi</subfield></datafield><datafield tag="028" ind1="5" ind2="2"><subfield code="a">/cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001773.pica</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)ELV057712875</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(ELSEVIER)S1077-3142(22)00037-6</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">630</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">640</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">610</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Dave, Ishan</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">TCLR: Temporal contrastive learning for video representation</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2022transfer abstract</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zzz</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">z</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zu</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR.</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local–local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global–local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4% (+5.1% increase over the previous best) top-1 accuracy on UCF101 and 52.9% (+5.4% increase) on HMDB51 action classification, and 56.2% (+11.7% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at https://github.com/DAVEISHAN/TCLR.</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">68T45</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">68T07</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">68T30</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Gupta, Rohit</subfield><subfield code="4">oth</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Rizve, Mamshad Nayeem</subfield><subfield code="4">oth</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Shah, Mubarak</subfield><subfield code="4">oth</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="n">Elsevier</subfield><subfield code="t">Editorial Board</subfield><subfield code="d">2016</subfield><subfield code="d">CVIU</subfield><subfield code="g">San Diego, Calif</subfield><subfield code="w">(DE-627)ELV019616708</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:219</subfield><subfield code="g">year:2022</subfield><subfield code="g">pages:0</subfield></datafield><datafield tag="856" ind1="4" ind2="0"><subfield code="u">https://doi.org/10.1016/j.cviu.2022.103406</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ELV</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-PHA</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">219</subfield><subfield code="j">2022</subfield><subfield code="h">0</subfield></datafield></record></collection>
|
| score |
7.400509 |