Eric Liu comments

Results 85 comments of


                                            Eric Liu

nvcc fatal : Unsupported gpu architecture 'compute_90'

You can try this way : https://blog.csdn.net/xunan003/article/details/90696412

Proactively let negative samples participate in training problems(主动让负样本参与训练问题)

可試試這個 https://github.com/eric612/MobileNet-YOLO/issues/206

How to generate train data

You can follow the [step ](https://hyzhan.github.io/2017/05/21/2017-05-21-%E7%94%A8SSD%E8%AE%AD%E7%BB%83%E8%87%AA%E5%B7%B1%E7%9A%84%E6%95%B0%E6%8D%AE%E9%9B%86(VOC2007%E6%A0%BC%E5%BC%8F)/) It is same as [ssd](https://github.com/weiliu89/caffe/tree/ssd)

Non merge_bn model

You can fine models here , https://github.com/eric612/MobileNet-YOLO/tree/master/models/yolov3 and see also , https://github.com/eric612/MobileNet-YOLO/issues/14

[Training issue]为什么在某些分辨率下，算法会对一类目标漏检？

不太理解，方便上傳log 嗎

[Training issue]为什么在某些分辨率下，算法会对一类目标漏检？

log 看起來沒什麼異常，但裡面沒有anchor的數值和test的結果，方便上傳嗎

do you have requirements.txt?

No , but I will add it in next commit .

encounter error when resuming from a trained .pth file

I removed the optimizer of 'model_best_my_train.pth.tar' , just only can be used in deploy or inference [Intermediate save code](https://github.com/eric612/Mobilenet-YOLO-Pytorch/blob/main/train.py#L150-L157) [model_best save code](https://github.com/eric612/Mobilenet-YOLO-Pytorch/blob/main/train.py#L360-L363)

problem about the Image processing time(600ms)

This is my code for estimating time ``` def inference_image(model, original_image,device): # Transforms transform_test = transforms.Compose([ transforms.Resize(size=(352,352), interpolation=2), transforms.ToTensor(), transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)), ]) # Transform image =...

problem about the Image processing time(600ms)

My device was GTX1080 Ti , your result was too strange . Did your gpu device was enabled ?