Official YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

[AlexeyAB]

Official YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

• paper: https://arxiv.org/abs/2207.02696

• source code - Pytorch (use to reproduce results): https://github.com/WongKinYiu/yolov7

---

Darknet cfg/weights file - currently tested for inference only:

• cfg: https://raw.githubusercontent.com/AlexeyAB/darknet/master/cfg/yolov7-tiny.cfg
• weights: https://github.com/AlexeyAB/darknet/releases/download/yolov4/yolov7-tiny.weights
• weights for fine-tuning: https://github.com/AlexeyAB/darknet/releases/download/yolov4/yolov7-tiny.conv.87

Test FPS on: https://github.com/AlexeyAB/darknet

• without NMS: darknet.exe detector demo cfg/coco.data yolov7-tiny.cfg yolov7-tiny.weights test.mp4 -benchmark

• with NMS: darknet.exe detector demo cfg/coco.data yolov7-tiny.cfg yolov7-tiny.weights test.mp4 -dont_show

---

YOLOv7 is more accurate and faster than YOLOv5 by 120% FPS, than YOLOX by 180% FPS, than Dual-Swin-T by 1200% FPS, than ConvNext by 550% FPS, than SWIN-L by 500% FPS, than PPYOLOE-X by 150% FPS.

YOLOv7 surpasses all known object detectors in both speed and accuracy in the range from 5 FPS to 160 FPS and has the highest accuracy 56.8% AP among all known real-time object detectors with 30 FPS or higher on GPU V100, batch=1.

• YOLOv7-e6 (55.9% AP, 56 FPS V100 b=1) by +500% FPS faster than SWIN-L C-M-RCNN (53.9% AP, 9.2 FPS A100 b=1)
• YOLOv7-e6 (55.9% AP, 56 FPS V100 b=1) by +550% FPS faster than ConvNeXt-XL C-M-RCNN (55.2% AP, 8.6 FPS A100 b=1)
• YOLOv7-w6 (54.6% AP, 84 FPS V100 b=1) by +120% FPS faster than YOLOv5-X6-r6.1 (55.0% AP, 38 FPS V100 b=1)
• YOLOv7-w6 (54.6% AP, 84 FPS V100 b=1) by +1200% FPS faster than Dual-Swin-T C-M-RCNN (53.6% AP, 6.5 FPS V100 b=1)
• YOLOv7x (52.9% AP, 114 FPS V100 b=1) by +150% FPS faster than PPYOLOE-X (51.9% AP, 45 FPS V100 b=1)
• YOLOv7 (51.2% AP, 161 FPS V100 b=1) by +180% FPS faster than YOLOX-X (51.1% AP, 58 FPS V100 b=1)

---

---

---

[ullrichthomas92]

I tried to train a custom dataset with yolov7-tiny and conv.87
Unfortunately it gets stuck at a loss above 100. Same configuration in yolov4-tiny works great.
Can you please help?

[ESJavadex]

Are you going to launch the full yolov7.weights and yolov7.cfg? Or are there any option to convert it from .pt?

[jhony2507]

Good Morning,
I'm using yolov7 to detect diseases in papaya, but the results are horrible. I have approximately 20k samples, divided into 8 diseases, the annotations are correct and I still get a max mAP of 34% (this after over 20,000 iterations). This result is slightly WORSE than yolov4, which achieves a mAP of 37%.
Does anyone have an idea what could be wrong?
*** With the efficientDet-d3 I get a mAP of 65% for the same base (I don't have the computational capacity to use the D6,D7..)

Attached are the yolov4/v7 configuration files and sample images

yolov4-custom_cfg.txt
yolov7-papaya_cfg.txt

[developer239]

@AlexeyAB @ESJavadex @jhony2507 I would also like to know what how do I get the weights file for yolov7? Can I convert .pt to .weights somehow? (I guess not)

How do I train .weights then? How is it different from training V3/V4? 🙏

I guess we can convert .pt to . onnx though. 🤔

[rishav1122]

I have the yolov7-w6 weights file in darknet format best.weights, Is there any way to convert it in pt. Also, is there a yolov7-W6 config file available (yolov7-w6.cfg)