End2end #61
End2end #61
Conversation
|
the code reference |
|
how is this different from your previous PR? |
wh = (wh * 2) ** 2 * self.anchor_grid[i].expand(shape) # the expand is error
return x if self.training else (torch.cat(z, 1), x)
return x if self.training else (z, ) if self.include_nms else (torch.cat(z, 1), x) |
add more tasks
update yolov7 prediction
|
Any reason to use EfficientNMS instead of BatchedNMSPlugin? https://github.com/NVIDIA/TensorRT/tree/main/plugin/batchedNMSPlugin |
|
Looking forward to it. Will 0.7 support yolov7 in some way? I guess all the changes from yolov5 apply to yolov7. |
Yep, we have the plan to support YOLOv7 at 0.7.0 release. |
|
@Linaom1214 Thanks! Could you provide Google Colab link how to do export and inference? |
Ok, here I provide a simple demo, the code training and inference are tested and the result is OK, and the end-to-end model export inference demo is provided. https://colab.research.google.com/drive/1JVC7Z_9163L7QL6SNxMyNeBCeFZpfqwc?usp=sharing |
fix nan of aux training WongKinYiu#250 (comment) @hudingding
|
Nice notebook. Do you know why the results are different though? One horse is missing. Maybe NMS parameters?
|
The precision fp32 maybe correct , you can do more experiments. Looking forward to your feedback |
|
Conf threshold was at 0.5 for TRT and 0.25 for Pytorch.
|
|
@Linaom1214 Thanks! Could you solve these 2 conflicts, that arose because of another merged pull request? https://github.com/WongKinYiu/yolov7/pull/114/files |
All test is right! Colab Demo |
* export end2end onnx model * fixbug * add web demo (WongKinYiu#58) * Update README.md * main code update yolov7-tiny deploy cfg * main code update yolov7-tiny training cfg * main code @liguagua752109150 WongKinYiu#33 (comment) * main code @albertfaromatics WongKinYiu#35 (comment) * main code update link * main code add custom hyp * main code update default activation function * main code update path * main figure add more tasks * main code update readme * main code update reparameterization * Update README.md * main code update readme * main code update aux training * main code update aux training * main code update aux training * main figure update yolov7 prediction * main code update readme * main code rename * main code rename * main code rename * main code rename * main code update readme * main code update visualization * main code fix gain for train_aux * main code update loss * main code update instance segmentation demo * main code update keypoint detection demo * main code update pose demo * main code update pose * main code update pose * main code update pose * main code update pose * main code update trace * Update README.md * main code fix ciou * main code fix nan of aux training WongKinYiu#250 (comment) @hudingding * support onnx to tensorrt convert (WongKinYiu#114) * fuse IDetect (WongKinYiu#148) * Fixes WongKinYiu#199 (WongKinYiu#203) * minor fix * resolve conflict * resolve conflict * resolve conflict * resolve conflict * resolve conflict * resolve * resolve * resolve * resolve Co-authored-by: AK391 <[email protected]> Co-authored-by: Alexey <[email protected]> Co-authored-by: Kin-Yiu, Wong <[email protected]> Co-authored-by: linghu8812 <[email protected]> Co-authored-by: Alexander <[email protected]> Co-authored-by: Ben Raymond <[email protected]> Co-authored-by: AlexeyAB84 <[email protected]>
export the onnx model include nms,than accelerate with tensorrt,
in 1080Ti the end2end model (yolov7-tiny) fps is 212, but the origin is 172.
onnx_graphsurgeonis needed