thograce / bgnet Goto Github PK

When I test I get the following error:

Traceback (most recent call last):
File "etest.py", line 30, in
image, gt, name = test_loader.load_data()
File "/data/zjb/camo/BGNet-master/utils/tdataloader.py", line 113, in load_data
gt = self.binary_loader(self.gts[self.index])
IndexError: list index out of range

What's wrong with this error?

What's the channels of corresponding GT? is it the same as the channels of extracted edge map?

Thank you~

I only saw the training set and test set in the project. Does the model training not require validation set? Looking forward to your reply.

Hello! I have now set the batchsize to 16 and the epoch of your source code to 25, but I set the epoch to 500, while I train and test, I find that the loss will continue to decrease when the epoch exceeds 25, so I think the model After the epoch exceeds 25, the effect should be better, so I tested it when epoch=126, but the test results are still different from the original paper. I would like to ask, isn't the lower the loss, the better the effect of the model? ?

i see that the input channel of the pretrained model is the gray image，why selecting the gray image to train the model ，if using the rgb image to train model ，the performance of the model will be better？when train the stereo matching model ，which image should be used，the gray image or rgb image？ thank you

I want to ask when the source code will be released. Thank you

How was the object-related edge exploration example for Image 7 in your paper generated?

i run etrain.py . the loss is nan.

I have a question about the number of pictures in the test dataset，test image is not 4k in cod10k??I found only 2026 pictures in your test dataset

Where to find the training set of edge image

Thanks for your nice work.
When I use the eval.py file to evaluate a pre-computed map for another model, I get the following error：

Traceback (most recent call last):
File "D:/1_code/1Net/eval.py", line 32, in
FM.step(pred=pred, gt=mask)
File "D:\Anaconda3\envs\pytorch\lib\site-packages\py_sod_metrics\sod_metrics.py", line 67, in step
adaptive_fm = self.cal_adaptive_fm(pred=pred, gt=gt)
File "D:\Anaconda3\envs\pytorch\lib\site-packages\py_sod_metrics\sod_metrics.py", line 84, in cal_adaptive_fm
area_intersection = binary_predcition[gt].sum()
IndexError: boolean index did not match indexed array along dimension 1; dimension is 700 but corresponding boolean dimension is 702

could you tell what problems in me my steps.

When I test I get the following error:
None of the test results matched the original image
The result is not related to the original image
For example:


What's the problem, please?

we dowmnoaded the BGNet.pth ，then run the etest.py to generate the predicted results, finally I run the eval.py. I get the final evalution metrics. But I found that the metric results which I tested has a big gap with that in the paper. could you tell what problems in me my steps. It is very importance for me! thanks your great contrubitions!

When I replay eval.py I get the following error when test the :
'maxFm': 0.7736841486165681}
63%|██████▎ | 2605/4121 [06:05<02:38, 9.59it/s][ WARN:[email protected]] global /io/opencv/modules/imgcodecs/src/loadsave.cpp (239) findDecoder imread_('./results/BGNet/NC4K/429.png'): can't open/read file: check file path/integrity
63%|██████▎ | 2606/4121 [06:05<03:32, 7.13it/s]
Traceback (most recent call last):
File "/ssdhome/qzb521/abc/BGNet/eval.py", line 30, in
FM.step(pred=pred, gt=mask)
File "/home/qzb521/miniconda3/envs/sinet/lib/python3.9/site-packages/py_sod_metrics/sod_metrics.py", line 65, in step
pred, gt = _prepare_data(pred, gt)
File "/home/qzb521/miniconda3/envs/sinet/lib/python3.9/site-packages/py_sod_metrics/sod_metrics.py", line 23, in _prepare_data
pred = pred / 255
TypeError: unsupported operand type(s) for /: 'NoneType' and 'int'

It maybe not generate the predict image before eval it? What's wrong with this error?

您好，关于多卡的训练，我修改了两处地方，包括如下：
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3"
def train(train_loader, model, optimizer, epoch):
model = torch.nn.DataParallel(model, device_ids=[0, 1, 2]).cuda() #添加
model.train()
[训练后的模型经过测试发现结果全是灰色图，如下

想问下您是否知道出了什么问题呢，万分感谢

Thanks for sharing the code.
I don't know how you calculate the camouflaged object edge( ground truth: Ge) based on the camouflaged object mask. Because the edge ground truth will be used in the loss function.
So, can you tell me how to get the camouflaged object edge map?