Kyle Min, Jason J. Corso
Temporally localizing activities within untrimmed videos has been extensively studied in recent years. Despite recent advances, existing methods for weakly-supervised temporal activity localization struggle to recognize when an activity is not occurring. To address this issue, we propose a novel method named A2CL-PT. Two triplets of the feature space are considered in our approach: one triplet is used to learn discriminative features for each activity class, and the other one is used to distinguish the features where no activity occurs (i.e. background features) from activity-related features for each video. To further improve the performance, we build our network using two parallel branches which operate in an adversarial way: the first branch localizes the most salient activities of a video and the second one finds other supplementary activities from non-localized parts of the video. Extensive experiments performed on THUMOS14 and ActivityNet datasets demonstrate that our proposed method is effective. Specifically, the average mAP of IoU thresholds from 0.1 to 0.9 on the THUMOS14 dataset is significantly improved from 27.9% to 30.0%.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Video | THUMOS’14 | mAP IOU@0.1 | 61.2 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.2 | 56.1 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.3 | 48.1 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.4 | 39 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.5 | 30.1 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.6 | 19.2 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.7 | 10.6 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.8 | 4.8 | A2CL-PT |
| Video | THUMOS’14 | mAP IOU@0.9 | 1 | A2CL-PT |
| Video | THUMOS’14 | mAP@AVG(0.1:0.9) | 30 | A2CL-PT |
| Video | THUMOS 2014 | mAP@0.1:0.5 | 46.9 | A2CL-PT |
| Video | THUMOS 2014 | mAP@0.1:0.7 | 37.8 | A2CL-PT |
| Video | THUMOS 2014 | mAP@0.5 | 30.1 | A2CL-PT |
| Video | THUMOS14 | avg-mAP (0.1-0.5) | 46.9 | A2CL-PT |
| Video | THUMOS14 | avg-mAP (0.1:0.7) | 37.8 | A2CL-PT |
| Video | THUMOS14 | avg-mAP (0.3-0.7) | 30.6 | A2CL-PT |
| Video | THUMOS’14 | mAP@0.5 | 30.1 | A2CL-PT |
| Video | ActivityNet-1.3 | mAP@0.5 | 36.8 | A2CL-PT |
| Video | ActivityNet-1.3 | mAP@0.5:0.95 | 22.5 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.1 | 61.2 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.2 | 56.1 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.3 | 48.1 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.4 | 39 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.5 | 30.1 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.6 | 19.2 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.7 | 10.6 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.8 | 4.8 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP IOU@0.9 | 1 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP@AVG(0.1:0.9) | 30 | A2CL-PT |
| Temporal Action Localization | THUMOS 2014 | mAP@0.1:0.5 | 46.9 | A2CL-PT |
| Temporal Action Localization | THUMOS 2014 | mAP@0.1:0.7 | 37.8 | A2CL-PT |
| Temporal Action Localization | THUMOS 2014 | mAP@0.5 | 30.1 | A2CL-PT |
| Temporal Action Localization | THUMOS14 | avg-mAP (0.1-0.5) | 46.9 | A2CL-PT |
| Temporal Action Localization | THUMOS14 | avg-mAP (0.1:0.7) | 37.8 | A2CL-PT |
| Temporal Action Localization | THUMOS14 | avg-mAP (0.3-0.7) | 30.6 | A2CL-PT |
| Temporal Action Localization | THUMOS’14 | mAP@0.5 | 30.1 | A2CL-PT |
| Temporal Action Localization | ActivityNet-1.3 | mAP@0.5 | 36.8 | A2CL-PT |
| Temporal Action Localization | ActivityNet-1.3 | mAP@0.5:0.95 | 22.5 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.1 | 61.2 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.2 | 56.1 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.3 | 48.1 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.4 | 39 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.5 | 30.1 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.6 | 19.2 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.7 | 10.6 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.8 | 4.8 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP IOU@0.9 | 1 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP@AVG(0.1:0.9) | 30 | A2CL-PT |
| Zero-Shot Learning | THUMOS 2014 | mAP@0.1:0.5 | 46.9 | A2CL-PT |
| Zero-Shot Learning | THUMOS 2014 | mAP@0.1:0.7 | 37.8 | A2CL-PT |
| Zero-Shot Learning | THUMOS 2014 | mAP@0.5 | 30.1 | A2CL-PT |
| Zero-Shot Learning | THUMOS14 | avg-mAP (0.1-0.5) | 46.9 | A2CL-PT |
| Zero-Shot Learning | THUMOS14 | avg-mAP (0.1:0.7) | 37.8 | A2CL-PT |
| Zero-Shot Learning | THUMOS14 | avg-mAP (0.3-0.7) | 30.6 | A2CL-PT |
| Zero-Shot Learning | THUMOS’14 | mAP@0.5 | 30.1 | A2CL-PT |
| Zero-Shot Learning | ActivityNet-1.3 | mAP@0.5 | 36.8 | A2CL-PT |
| Zero-Shot Learning | ActivityNet-1.3 | mAP@0.5:0.95 | 22.5 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.1 | 61.2 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.2 | 56.1 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.3 | 48.1 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.4 | 39 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.5 | 30.1 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.6 | 19.2 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.7 | 10.6 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.8 | 4.8 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP IOU@0.9 | 1 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP@AVG(0.1:0.9) | 30 | A2CL-PT |
| Action Localization | THUMOS 2014 | mAP@0.1:0.5 | 46.9 | A2CL-PT |
| Action Localization | THUMOS 2014 | mAP@0.1:0.7 | 37.8 | A2CL-PT |
| Action Localization | THUMOS 2014 | mAP@0.5 | 30.1 | A2CL-PT |
| Action Localization | THUMOS14 | avg-mAP (0.1-0.5) | 46.9 | A2CL-PT |
| Action Localization | THUMOS14 | avg-mAP (0.1:0.7) | 37.8 | A2CL-PT |
| Action Localization | THUMOS14 | avg-mAP (0.3-0.7) | 30.6 | A2CL-PT |
| Action Localization | THUMOS’14 | mAP@0.5 | 30.1 | A2CL-PT |
| Action Localization | ActivityNet-1.3 | mAP@0.5 | 36.8 | A2CL-PT |
| Action Localization | ActivityNet-1.3 | mAP@0.5:0.95 | 22.5 | A2CL-PT |
| Weakly Supervised Action Localization | THUMOS 2014 | mAP@0.1:0.5 | 46.9 | A2CL-PT |
| Weakly Supervised Action Localization | THUMOS 2014 | mAP@0.1:0.7 | 37.8 | A2CL-PT |
| Weakly Supervised Action Localization | THUMOS 2014 | mAP@0.5 | 30.1 | A2CL-PT |
| Weakly Supervised Action Localization | THUMOS14 | avg-mAP (0.1-0.5) | 46.9 | A2CL-PT |
| Weakly Supervised Action Localization | THUMOS14 | avg-mAP (0.1:0.7) | 37.8 | A2CL-PT |
| Weakly Supervised Action Localization | THUMOS14 | avg-mAP (0.3-0.7) | 30.6 | A2CL-PT |
| Weakly Supervised Action Localization | THUMOS’14 | mAP@0.5 | 30.1 | A2CL-PT |
| Weakly Supervised Action Localization | ActivityNet-1.3 | mAP@0.5 | 36.8 | A2CL-PT |
| Weakly Supervised Action Localization | ActivityNet-1.3 | mAP@0.5:0.95 | 22.5 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.1 | 61.2 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.2 | 56.1 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.3 | 48.1 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.4 | 39 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.5 | 30.1 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.6 | 19.2 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.7 | 10.6 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.8 | 4.8 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP IOU@0.9 | 1 | A2CL-PT |
| Weakly-supervised Temporal Action Localization | THUMOS’14 | mAP@AVG(0.1:0.9) | 30 | A2CL-PT |