pspnet-pytorch

pspnet-pytorch copied to clipboard

解冻之后mIou飞速下降，并且训练完100epoch效果还是很差，远达不到作者的mIou

xhxmssb

Traceback (most recent call last): File "/home/jiaorunnong/pspnet-pytorch-master/get_miou.py", line 60, in hist, IoUs, PA_Recall, Precision = compute_mIoU(gt_dir, pred_dir, image_ids, num_classes, name_classes) # 执行计算mIoU的函数 File "/home/jiaorunnong/pspnet-pytorch-master/utils/utils_metrics.py", line 118, in compute_mIoU hist +=...

LeonardJiao

转化onnx是算子不支持，这个有什么办法吗

5

RuntimeError: Unsupported: ONNX export of operator adaptive_avg_pool2d, since output size is not factor of input size. Please feel free to request support or submit a pull request on PyTorch GitHub.

chen-del

Ttingyyy

C:\Users\97328\.conda\envs\torch\python.exe "E:\A_Deep learning model\pspnet-txx-yck\predict.py" E:\A_Deep learning model\pspnet-txx-yck\logs\lr0.0001\best_epoch_weights.pth model, and classes loaded. Configurations: ---------------------------------------------------------------------- | keys | values| ---------------------------------------------------------------------- | model_path | E:\A_Deep learning model\pspnet-txx-yck\logs\lr0.0001\best_epoch_weights.pth| | num_classes | 2| | backbone...

t973288913

代码查看灰度值全是0，请问是什么原因哇

hhzxy

博主你好，我在训练自己数据集时出现fscore一直很高在91左右，但是在训练是评估的miou却很低，只有9。训练的数据集是cityspaces， 在train文件中修改了类别，不使用预训练权重，请问还需要写明具体 类别吗，训练文件中好像并没有找到修改的地方，还是有其余部分有设置错误嘛

zoom521241

请问这个是哪里出了问题，我没改代码

3

File "/root/pspnet-pytorch-master/utils/callbacks.py", line 135, in get_miou_png pr = F.softmax(pr.permute(1,2,0),dim = -1).cpu().numpy() RuntimeError: number of dims don't match in permute

425her0

训练集，测试集

3

楼主我想问一下，整个训练过程仅用VOCdevkit作为训练集，未用到验证集么

njust-lyy

unet里的resnet50.py和pspnet里的resnet50.py不一样嘛？

1

https://github.com/bubbliiiing/pspnet-pytorch/blob/f4aba6143a42b3f8c6dfdbe7a0858410227b1a29/nets/resnet.py#L64 这里的卷积核大小是3，但是unet那个repository里的resnet50卷积核大小是7？ https://github.com/bubbliiiing/unet-pytorch/blob/8ab373232c5c3a1877f3e84e6f5d97404089c20f/nets/resnet.py#L109 这个有什么讲究吗？谢谢

chenvvenxin