- YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
- Implementation on Github
Performance
Breaking Down YOLOv7
Inference on Traffic Video
| Video | Model | Image Size | # frames | elapsed time | fps |
|---|---|---|---|---|---|
| Isanpo Interchange Day | YOLOv7 | 640px | 191 frames | 56 fps | 3.41 sec |
| Isanpo Interchange Night | YOLOv7 | 640px | 194 frames | 56 fps | 3.43 sec |
| Rainy Seoul Street | YOLOv7-E6E | 1280px | 4606 frames | 16 fps | 287.05 sec |
Isanpo Interchange CCTV
Rainy Seoul Street
original video from @SeoulWalker