custom classes about m2det

[Maxfashko]

@ufohuang98 I could not fix this problem. It seems the developers need to pay attention to this problem, and explain why this is happening and how to solve it. @dshahrokhian we can hope to get clear explanations from you about this problem? Or do you prefer to leave us alone with this problem? :)

[Davis-love-AI]

@dshahrokhian @Maxfashko I also meet the the problem.....and I dont know how to solve.
I retrain VOC0712, and get good results, but when train my dataset for one class, get random result.

[lonl]

@dshahrokhian @Maxfashko @deqiangwang @Xiehuaiqi @ufohuang98 @rw1995 maybe you should adjust the weights between "loss_l" and "loss_c", you could find "loss = loss_l + loss_c" in "train.py" of this project, and you can introduce a parameter lambda to change the loss, like "loss = loss_l + 0.1*loss_c", finally, you may get the correct results~

[rw1995]

pos:(4.4,-49.1,156.7,199.6), ids:person, score:1.000
pos:(119.9,-47.5,279.2,201.6), ids:person, score:1.000
pos:(1748.8,-38.1,1903.9,212.6), ids:person, score:1.000
pos:(18.3,485.6,142.4,695.9), ids:person, score:1.000
pos:(70.4,538.3,208.6,767.5), ids:person, score:1.000
pos:(185.5,525.4,332.5,771.1), ids:person, score:1.000
pos:(302.0,515.1,455.5,770.1), ids:person, score:1.000
pos:(122.0,592.4,278.3,839.3), ids:person, score:1.000
pos:(19.2,632.0,167.2,871.7), ids:person, score:1.000
pos:(235.8,613.2,399.0,877.5), ids:person, score:1.000
pos:(355.3,608.2,513.1,884.7), ids:person, score:1.000
pos:(480.6,646.6,637.0,931.0), ids:person, score:1.000
pos:(1191.1,629.0,1369.8,917.5), ids:person, score:1.000
pos:(602.5,688.8,752.3,965.3), ids:person, score:1.000
pos:(727.8,688.2,874.3,966.8), ids:person, score:1.000
pos:(850.8,681.3,1000.4,966.4), ids:person, score:1.000
pos:(958.8,673.6,1123.7,962.8), ids:person, score:1.000
pos:(1072.2,670.2,1242.4,957.1), ids:person, score:1.000
pos:(1310.2,669.1,1483.1,953.0), ids:person, score:1.000
pos:(1438.5,673.7,1605.4,966.8), ids:person, score:1.000
pos:(62.7,733.4,233.6,950.0), ids:person, score:1.000
pos:(299.5,729.1,474.0,963.3), ids:person, score:1.000
pos:(412.1,729.8,578.8,992.9), ids:person, score:1.000
pos:(1574.7,737.8,1724.7,990.1), ids:person, score:1.000
pos:(1748.5,732.4,1903.9,1004.1), ids:person, score:1.000
pos:(203.6,766.6,362.1,964.2), ids:person, score:1.000
pos:(785.6,755.0,941.1,1028.1), ids:person, score:1.000
pos:(530.8,772.5,697.8,1041.7), ids:person, score:1.000
pos:(654.1,772.0,817.2,1048.8), ids:person, score:1.000
pos:(901.4,767.8,1058.5,1053.1), ids:person, score:1.000
pos:(1020.6,767.3,1181.0,1053.5), ids:person, score:1.000
pos:(1142.7,776.6,1299.8,1046.9), ids:person, score:1.000
pos:(1269.4,783.5,1420.9,1040.9), ids:person, score:1.000
pos:(1403.6,792.3,1547.0,1031.9), ids:person, score:1.000
pos:(1522.8,705.5,1652.6,1089.5), ids:person, score:1.000
pos:(1649.7,791.9,1779.5,1030.4), ids:person, score:1.000
pos:(-25.3,852.3,136.6,1073.3), ids:person, score:1.000
pos:(112.4,779.8,283.6,1116.6), ids:person, score:1.000
pos:(369.2,794.7,514.0,1157.2), ids:person, score:1.000
pos:(1224.7,771.4,1345.4,1168.1), ids:person, score:1.000
pos:(241.4,818.6,402.8,1182.6), ids:person, score:1.000

Confidence 1.0 on all objects - this is at least strange :)

can u help me? when i run the demo.py following the progress ,but have a wrong .
I do sh make.sh and python demo.py -c=configs/m2det512_vgg.py -m=weights/m2det512_vgg.pth --show
but the result is

The Anchor info:
{'variance': [0.1, 0.2], 'min_sizes': [30.72, 76.8, 168.96, 261.12, 353.28, 445.44], 'aspect_ratios': [[2, 3], [2, 3], [2, 3], [2, 3], [2, 3], [2, 3]], 'clip': True, 'feature_maps': [64, 32, 16, 8, 4, 2], 'min_dim': 512, 'max_sizes': [76.8, 168.96, 261.12, 353.28, 445.44, 537.6], 'steps': [8, 16, 32, 64, 128, 256]}
===> Constructing M2Det model
Loading resume network...
===> Finished constructing and loading model
<class 'm2det.M2Det'>
THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=663 error=11 : invalid argument
Traceback (most recent call last):
File "demo.py", line 113, in
out = net(img)
File "/home/csy/anaconda3/envs/pytorch0.4.1/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/csy/M2Det/m2det.py", line 106, in forward
x = self.basek
File "/home/csy/anaconda3/envs/pytorch0.4.1/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/csy/anaconda3/envs/pytorch0.4.1/lib/python3.5/site-packages/torch/nn/modules/conv.py", line 301, in forward
self.padding, self.dilation, self.groups)
RuntimeError: cuda runtime error (11) : invalid argument at /pytorch/aten/src/THC/THCGeneral.cpp:663

how can you run the demo.py? do you modified some locations?I will be grateful to you
@Maxfashko

ok，i solve it .just my environment about pytorch is wrong .

[ufohuang98]

@Maxfashko

I have the same issue ,would you please tell me how did you fixed it?
I use pytorch 0.4.1, coco dota format and trained 26 epoch.

[dshahrokhian]

Just a random thought, did not try it myself: num_classes=2, since __background__ is already a class.

[Maxfashko]

@dshahrokhian thank, it helped.
I have one more question. My model trained 10 epoch (600 image). I run demo.py script and see than bbox aligned randomly over image. What shall i do? Is this a normal situation or i need to train more time?

[Maxfashko]

pos:(4.4,-49.1,156.7,199.6), ids:person, score:1.000
pos:(119.9,-47.5,279.2,201.6), ids:person, score:1.000
pos:(1748.8,-38.1,1903.9,212.6), ids:person, score:1.000
pos:(18.3,485.6,142.4,695.9), ids:person, score:1.000
pos:(70.4,538.3,208.6,767.5), ids:person, score:1.000
pos:(185.5,525.4,332.5,771.1), ids:person, score:1.000
pos:(302.0,515.1,455.5,770.1), ids:person, score:1.000
pos:(122.0,592.4,278.3,839.3), ids:person, score:1.000
pos:(19.2,632.0,167.2,871.7), ids:person, score:1.000
pos:(235.8,613.2,399.0,877.5), ids:person, score:1.000
pos:(355.3,608.2,513.1,884.7), ids:person, score:1.000
pos:(480.6,646.6,637.0,931.0), ids:person, score:1.000
pos:(1191.1,629.0,1369.8,917.5), ids:person, score:1.000
pos:(602.5,688.8,752.3,965.3), ids:person, score:1.000
pos:(727.8,688.2,874.3,966.8), ids:person, score:1.000
pos:(850.8,681.3,1000.4,966.4), ids:person, score:1.000
pos:(958.8,673.6,1123.7,962.8), ids:person, score:1.000
pos:(1072.2,670.2,1242.4,957.1), ids:person, score:1.000
pos:(1310.2,669.1,1483.1,953.0), ids:person, score:1.000
pos:(1438.5,673.7,1605.4,966.8), ids:person, score:1.000
pos:(62.7,733.4,233.6,950.0), ids:person, score:1.000
pos:(299.5,729.1,474.0,963.3), ids:person, score:1.000
pos:(412.1,729.8,578.8,992.9), ids:person, score:1.000
pos:(1574.7,737.8,1724.7,990.1), ids:person, score:1.000
pos:(1748.5,732.4,1903.9,1004.1), ids:person, score:1.000
pos:(203.6,766.6,362.1,964.2), ids:person, score:1.000
pos:(785.6,755.0,941.1,1028.1), ids:person, score:1.000
pos:(530.8,772.5,697.8,1041.7), ids:person, score:1.000
pos:(654.1,772.0,817.2,1048.8), ids:person, score:1.000
pos:(901.4,767.8,1058.5,1053.1), ids:person, score:1.000
pos:(1020.6,767.3,1181.0,1053.5), ids:person, score:1.000
pos:(1142.7,776.6,1299.8,1046.9), ids:person, score:1.000
pos:(1269.4,783.5,1420.9,1040.9), ids:person, score:1.000
pos:(1403.6,792.3,1547.0,1031.9), ids:person, score:1.000
pos:(1522.8,705.5,1652.6,1089.5), ids:person, score:1.000
pos:(1649.7,791.9,1779.5,1030.4), ids:person, score:1.000
pos:(-25.3,852.3,136.6,1073.3), ids:person, score:1.000
pos:(112.4,779.8,283.6,1116.6), ids:person, score:1.000
pos:(369.2,794.7,514.0,1157.2), ids:person, score:1.000
pos:(1224.7,771.4,1345.4,1168.1), ids:person, score:1.000
pos:(241.4,818.6,402.8,1182.6), ids:person, score:1.000

Confidence 1.0 on all objects - this is at least strange :)

[rw1995]

pos:(4.4,-49.1,156.7,199.6), ids:person, score:1.000
pos:(119.9,-47.5,279.2,201.6), ids:person, score:1.000
pos:(1748.8,-38.1,1903.9,212.6), ids:person, score:1.000
pos:(18.3,485.6,142.4,695.9), ids:person, score:1.000
pos:(70.4,538.3,208.6,767.5), ids:person, score:1.000
pos:(185.5,525.4,332.5,771.1), ids:person, score:1.000
pos:(302.0,515.1,455.5,770.1), ids:person, score:1.000
pos:(122.0,592.4,278.3,839.3), ids:person, score:1.000
pos:(19.2,632.0,167.2,871.7), ids:person, score:1.000
pos:(235.8,613.2,399.0,877.5), ids:person, score:1.000
pos:(355.3,608.2,513.1,884.7), ids:person, score:1.000
pos:(480.6,646.6,637.0,931.0), ids:person, score:1.000
pos:(1191.1,629.0,1369.8,917.5), ids:person, score:1.000
pos:(602.5,688.8,752.3,965.3), ids:person, score:1.000
pos:(727.8,688.2,874.3,966.8), ids:person, score:1.000
pos:(850.8,681.3,1000.4,966.4), ids:person, score:1.000
pos:(958.8,673.6,1123.7,962.8), ids:person, score:1.000
pos:(1072.2,670.2,1242.4,957.1), ids:person, score:1.000
pos:(1310.2,669.1,1483.1,953.0), ids:person, score:1.000
pos:(1438.5,673.7,1605.4,966.8), ids:person, score:1.000
pos:(62.7,733.4,233.6,950.0), ids:person, score:1.000
pos:(299.5,729.1,474.0,963.3), ids:person, score:1.000
pos:(412.1,729.8,578.8,992.9), ids:person, score:1.000
pos:(1574.7,737.8,1724.7,990.1), ids:person, score:1.000
pos:(1748.5,732.4,1903.9,1004.1), ids:person, score:1.000
pos:(203.6,766.6,362.1,964.2), ids:person, score:1.000
pos:(785.6,755.0,941.1,1028.1), ids:person, score:1.000
pos:(530.8,772.5,697.8,1041.7), ids:person, score:1.000
pos:(654.1,772.0,817.2,1048.8), ids:person, score:1.000
pos:(901.4,767.8,1058.5,1053.1), ids:person, score:1.000
pos:(1020.6,767.3,1181.0,1053.5), ids:person, score:1.000
pos:(1142.7,776.6,1299.8,1046.9), ids:person, score:1.000
pos:(1269.4,783.5,1420.9,1040.9), ids:person, score:1.000
pos:(1403.6,792.3,1547.0,1031.9), ids:person, score:1.000
pos:(1522.8,705.5,1652.6,1089.5), ids:person, score:1.000
pos:(1649.7,791.9,1779.5,1030.4), ids:person, score:1.000
pos:(-25.3,852.3,136.6,1073.3), ids:person, score:1.000
pos:(112.4,779.8,283.6,1116.6), ids:person, score:1.000
pos:(369.2,794.7,514.0,1157.2), ids:person, score:1.000
pos:(1224.7,771.4,1345.4,1168.1), ids:person, score:1.000
pos:(241.4,818.6,402.8,1182.6), ids:person, score:1.000

Confidence 1.0 on all objects - this is at least strange :)

can u help me? when i run the demo.py following the progress ,but have a wrong .
I do sh make.sh and python demo.py -c=configs/m2det512_vgg.py -m=weights/m2det512_vgg.pth --show
but the result is

The Anchor info:
{'variance': [0.1, 0.2], 'min_sizes': [30.72, 76.8, 168.96, 261.12, 353.28, 445.44], 'aspect_ratios': [[2, 3], [2, 3], [2, 3], [2, 3], [2, 3], [2, 3]], 'clip': True, 'feature_maps': [64, 32, 16, 8, 4, 2], 'min_dim': 512, 'max_sizes': [76.8, 168.96, 261.12, 353.28, 445.44, 537.6], 'steps': [8, 16, 32, 64, 128, 256]}
===> Constructing M2Det model
Loading resume network...
===> Finished constructing and loading model
<class 'm2det.M2Det'>
THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=663 error=11 : invalid argument
Traceback (most recent call last):
File "demo.py", line 113, in
out = net(img)
File "/home/csy/anaconda3/envs/pytorch0.4.1/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/csy/M2Det/m2det.py", line 106, in forward
x = self.basek
File "/home/csy/anaconda3/envs/pytorch0.4.1/lib/python3.5/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/csy/anaconda3/envs/pytorch0.4.1/lib/python3.5/site-packages/torch/nn/modules/conv.py", line 301, in forward
self.padding, self.dilation, self.groups)
RuntimeError: cuda runtime error (11) : invalid argument at /pytorch/aten/src/THC/THCGeneral.cpp:663

how can you run the demo.py? do you modified some locations?I will be grateful to you
@Maxfashko

[dshahrokhian]

Did it work with pytorch 0.4?

[Maxfashko]

@dshahrokhian no, i use pytorch 1.0

[dshahrokhian]

I meant @rw1995, as it seems to work for him now, and the repo README says to use 0.4.1

[Xiehuaiqi]

I meet the same question to train one class.if you solve ,please help me

[Maxfashko]

@Xiehuaiqi I stopped trying to make the code work that is obviously not working :)
The author chose to ignore our questions. Either @dshahrokhian is not competent, or my assumption is true - the code is not working :)

[taehyunzzz]

@lonl Hi, you mentioned weighting the 2 loss values as a possible resolution. Did this work out for you?

[lonl]

@taehyunzzz, I have conducted this way to try to solve the problem, however, some pics still have many error bounding boxes, and maybe this way could not work ~