FileNotFoundError: [Errno 2] No such file or directory: 'baseline-resnet50dilated-ppm_deepsup/encoder_epoch_20.pth'

[IceStreams]

Hello，thanks for your work. When I run train.py, error FileNotFoundError: [Errno 2] No such file or directory: 'Baseline - resnet50dilating-ppm_deepsup /encoder_epoch_20.pth'.Where can I find this file

[vidit98]

Hi, thank you for your interest in the work. You can find the model here.

[IceStreams]

Hello, thanks for your work. I have downloaded the ADE20K data set as requested, and placed all the training images (such as ADE_train_00000001.jpg) in data/ADEChallengeData2016/images/training and ground truth of segmentation (such as ADE_train_00000001_seg.png) in data/ADEChallengeData2016/annotations/training. but when I run the train.py. It report an error as follows: File "E:\Experiment\graphconv-master\model.py", line 60, in forward acc = self.pixel_acc(pred, feed_dict['seg_label'].cuda(0)) RuntimeError: CUDA error: device-side assert triggered. I looked for this kind of error on the Internet, which should be caused by the tag crossing. Then I printed out the maximum value of the variable "feed_dict['seg_label'].cuda()", and found that the maximum value was lager than the number of class (150). Is it because my data of ground truth is wrong，Have you ever come across the same problem?

[vidit98]

Yes, I also faced this issue once, it is generally border case which causes problem for example numbering in ground truth starts from 0 and you have started it from 1. Although it should not have occurred. What is the maximum value that you are getting when you print it?

[IceStreams]

Because the training samples in your code are randomly selected, the maximum value printed will be different each time. But they're all larger than 150, like 184,169,157 etc. How did you solve this problem?

[IceStreams]

I compare the ADE_train_00000001.jpg file I downloaded from the ADE20K dataset with the ADE_train_00000001.jpg file that exists in your code. I find that the two are different. Does this mean that the data set I downloaded is wrong? I downloaded the data set from here . it is the url you provided.

[vidit98]

There can be a possibility that the dataset has changed, you can check whether they have increased the number of classes from 150. Another thing you can check is that you are putting segmentation images in the ground truth and not the part segmented image.