Mask Predict

[DannyChen1994]

Hello, I have installed the environment and tested try_test_net.py code can be run. I just started out and don't have a deep understanding of code. I haven't seen any code about mask visualization. May I ask if using the forward function in try_demo_img. py? Can you provide examples for guidance?

[youngnuaa]

the code specification is not too good. try_demo_img.py is used to show the result of model prediction, try_test_net.py is used to evaluate the map of the model, and train_net.py is used to train the model.

[DannyChen1994]

Thank you for your guidance. I have run through the previous code and successfully trained my dataset. But I found a problem when I evaluated the "model_final.pth" trained on my power-line synthesis dataset on the author's wire dataset. May I ask what may have caused this？Is it caused by not detecting the target？

[youngnuaa]

I suggest print "cate_scores" to show the result.

[DannyChen1994]

Yes, I found that for graphs that have not been split into instances, the code will report an error, and 'cate_cores' is empty data. In addition, I found that when the input image is not square, such as the shape at 16:9, there will be a mismatch between the mask and the actual object, as shown in the following figure. When I modify the 'Base SpareInst. yaml' in the 'configs' folder (MAX_SIZE_TEST: 864). There is a mismatch in the mask in the high direction. May I ask what the reason is and how can we solve it?

[DannyChen1994]

I made the following modification MIN_SIZE_TEST: 1024 MAX_SIZE_TEST: 1024 The wiping film is normal, and the problem described above is solved. Thank you！