jingdao / ir_detection Goto Github PK

Hello , Thanks for your share code. I follow your introduction to run the train.py but get the result and I want to ask you why this happen and how to solve this problem? Thank you
'''
Epoch 0 loss nan(nan+1.68) prc 0.03 rcl 0.87
Epoch 1 loss nan(nan+0.61) prc 0.05 rcl 0.83
Epoch 2 loss nan(nan+0.39) prc 0.14 rcl 0.89
Epoch 3 loss nan(nan+0.29) prc 0.19 rcl 0.84
Epoch 4 loss nan(nan+0.22) prc 0.20 rcl 0.68
Epoch 5 loss nan(nan+0.23) prc 0.20 rcl 0.67
Epoch 6 loss nan(nan+0.25) prc 0.15 rcl 0.94
'''

Hey!
imresize is deprecated in SciPy 1.0.0, and will be removed in 1.3.0 can you please tell an alternative for:
if imscale!=imsize:
image_np = scipy.misc.imresize(image_np, size=(imsize, imsize), interp='bilinear')
annotation = scipy.misc.imresize(annotation, size=(imsize, imsize), interp='bilinear')

in process_record.py. Thank you!

So I tried to run annotation.py with my own dataset and with my custom margin, after I finished to annotate and closed the annotation tool, I saw my label1.png and the rest of my labeled images were just black, no segmented label from annotation.py was saved. Was it normal or did I do it wrong? Thanks

How can I run test on my own images not includes your datasets and labels?

Dear author, thank you very much for your sharing, and I am also engaged in related research. But we don’t have a relevant data set. Can you publish your data set?

Hi, thanks for open-sourcing this repo.

Is there a script to use the code on unknown dataset with no labels to test the model ?
The code test.py needs the label images to test the model...

I tried modifying it (in a new inference.py script) but the prediction line does not return the same output as in test.py:
result = sess.run(
[net.logits_tf],
{net.input_pl:input_images})

In that case, what would be the changes in the predicted_softmax line :
predicted_softmax = scipy.special.softmax(result[0,:,:,:],axis=-1)[:,:,1]

Thanks for any answers that may help !

Dear author, thank you very much for sharing on github, but since I am not familiar with the tensorflow framework, can you provide the pytorch version of the code? ? ? If not, can the "train.py" and "test.py" in github be replaced with pytorch versions?