training error

[ladyxuxu]

hi，do you have met this error :RuntimeError: cannot perform reduction function max on tensor with no elements because the operation does not have an identity

and Would you mind share me the version of pytorch you use ?

[cowarder]

You need to create a folder 'imgs' in 'simulator_dataset' to put simulator images mentioned in the paper. The newest stable pytorch GPU version is Ok.

[ladyxuxu]

yes , Mistakes occur during training , after train iterations ,it random happen. could mind provide your e-mail ,and i attach the picture for reference

[ladyxuxu]

[ladyxuxu]

i track , the y_true[b, ..., 0:4] has value ,but true_box = y_true[b, ..., 0:4][object_mask_bool[b, ..., 0]] always empty ,is 0 .

[cowarder]

This happens when the image doesn't exist, the dataset lacks some images. You can print the image path to check which of the images doesn't exist and delete it in the txt file.

[ladyxuxu]

as you say ,this is my test result ，img exist meanwhile lableinformation exist ,so i dont understand ,what happened

[cowarder]

If you meet a randomly happening error, you should pay more attention to error samples. If you check the error image, you'll find there isn't any object(car) in the image.

[ladyxuxu]

yes , i open the picture and find no car in pic, and i want to know "img_r_805.png 0,189,1,202,0,1,201,0,202,0,189,1,192" no car exist in pic ,What do these labels represent?

[cowarder]

Not sure, maybe just some dirty samples.