apxl
apxl
@philferriere Thank you for your discussion. Is there any updated results using the correct fuction(i.e. query_points_on_grid = batched_grid + flow)? I noticed that the code in master branch is still...
@Banhalmi Have you solved the problem? I met the same problem.
> thank you. [email protected] 发件人： jerryho-quanta 发送时间： 2019-11-08 09:58 收件人： Lam1360/YOLOv3-model-pruning 抄送： MJyang; Comment 主题： Re: [Lam1360/YOLOv3-model-pruning] COCO 测试减枝map 很低? (#48) I use coco for pruning....also map is very low...
@CF2220160244 @luluvi Hi, I apply the network-slimming approach on mobilenetv2 [MobileNet-v2-pruning](https://github.com/wlguan/MobileNet-v2-pruning). The code is in pytorch and organized like this project. I hope this could help.
@LittleBoy7 @LiJunnan1992 Hi, I reimplemented kl-loss in with my own project [Stronger-yolo-pytorch](https://github.com/wlguan/Stronger-yolo-pytorch). You may find some help here.
@zachteed same problem, do you have any solution? Or which CUDA version is required?
建议使用c++版本的代码进行测试，如果能正常出结果说明不是模型的问题。
我没有直接尝试过用这个tf版本训练导出，建议你使用原版代码训练，用我修改后的脚本导出。pb模型预测错误可以逐层debug。
@betterhalfwzm Hi, the author did a great job and I reimplemented it with pytorch [MobileNet-v2-pruning](https://github.com/wlguan/MobileNet-v2-pruning), which is more readable than caffe version.