Undefined functions appear when running train_CCAM_VOC12.py？

[liaochuanlin]

File "/home/lcl/BECO/BECO-main/CCAM-master/WSSS/train_CCAM_VOC12.py", line 240, in
visualize_heatmap(args.tag, images.clone().detach(), ccam, 0, iteration)
NameError: name 'visualize_heatmap' is not defined

[Sierkinhane]

Hi, the visualize_heatmap() function was defined in WSSS/utils and please ensure to import it.

[liaochuanlin]

Thank you for your reply, but there are other problems.
Traceback (most recent call last):
File "/home/lcl/BECO/BECO-main/CCAM-master/WSSS/train_CCAM_VOC12.py", line 273, in
torch.save({'state_dict': model.module.state_dict() if (the_number_of_gpu > 1) else model.state_dict(),
File "/home/lcl/anaconda3/envs/BECO/lib/python3.10/site-packages/torch/serialization.py", line 440, in save
with _open_zipfile_writer(f) as opened_zipfile:
File "/home/lcl/anaconda3/envs/BECO/lib/python3.10/site-packages/torch/serialization.py", line 315, in _open_zipfile_writer
return container(name_or_buffer)
File "/home/lcl/anaconda3/envs/BECO/lib/python3.10/site-packages/torch/serialization.py", line 288, in init
super().init(torch._C.PyTorchFileWriter(str(name)))
RuntimeError: [enforce fail at inline_container.cc:365] . invalid file name: ./experiments/models/.pth

[Sierkinhane]

Hi, actually the program will automatically create the folder experiments/models/. You can also manually create them.

[liaochuanlin]

Traceback (most recent call last):
File "/home/lcl/BECO/BECO-main/CCAM-master/WSSS/inference_crf.py", line 98, in
cams = crf_inference_label(np.asarray(ori_image), cams, n_labels=keys.shape[0], t=args.crf_iteration)
File "/home/lcl/BECO/BECO-main/CCAM-master/WSSS/tools/ai/demo_utils.py", line 102, in crf_inference_label
d = dcrf.DenseCRF2D(w, h, n_labels)
AttributeError: module 'pydensecrf.densecrf' has no attribute 'DenseCRF2D'

[Sierkinhane]

Installing pydensecrf from source can address the issue.

[liaochuanlin]

I downloaded the whole folder and added it to wsss, but the above problem still occurred. Is it the wrong way to download the whole file?

[Sierkinhane]

Install using python3 setup.py install.

[liaochuanlin]

When running the evaluation：
/home/lcl/anaconda3/envs/BECO/bin/python3.10 /home/lcl/BECO/BECO-main/CCAM-master/WSSS/evaluate_using_background_cues.py --domain train --with_bg_cues --data_dir /home/lcl/BECO/BECO-main/CCAM-master/WSSS/data/VOC2012 True --bg_dir /home/lcl/BECO/BECO-main/CCAM-master/WSSS/experiments/predictions/CCAM_VOC12_MOCO@train@scale=0.5,1.0,1.5,2.0@t=0.3@ccam_inference_crf=10 True
usage: evaluate_using_background_cues.py [-h]
[--experiment_name EXPERIMENT_NAME]
[--domain DOMAIN]
[--threshold THRESHOLD]
[--predict_dir PREDICT_DIR]
[--gt_dir GT_DIR] [--logfile LOGFILE]
[--comment COMMENT] [--mode MODE]
[--max_th MAX_TH] [--bg_dir BG_DIR]
[--with_bg_cues WITH_BG_CUES]
evaluate_using_background_cues.py: error: argument --with_bg_cues: expected one argument

[liaochuanlin]

/home/lcl/anaconda3/envs/BECO/bin/python3.10 /home/lcl/BECO/BECO-main/CCAM-master/WSSS/evaluate_using_background_cues.py --domain train --with_bg_cues True --bg_dir /home/lcl/BECO/BECO-main/CCAM-master/WSSS/experiments/predictions/CCAM_VOC12_MOCO@train@scale=0.5,1.0,1.5,2.0@t=0.3@ccam_inference_crf=10
Th=0.05, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.10, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.15, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.20, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.25, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.30, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.35, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.40, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.45, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.50, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.55, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.60, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.65, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.70, mIoU=3.599%, FP=0.0000, FN=1.0000
Th=0.75, mIoU=3.599%, FP=0.0000, FN=1.0000
Best Th=0.05, mIoU=3.599%
The end result is very bad.

[liaochuanlin]

type=str)

parser.add_argument('--experiment_name', default='CCAM_VOC12_MOCO@train@scale=0.5,1.0,1.5,2.0@t=0.3@ccam_inference_crf=10',
type=str)
parser.add_argument("--domain", default='train', type=str)
parser.add_argument("--threshold", default=None, type=float)

parser.add_argument("--predict_dir", default='', type=str)
parser.add_argument('--gt_dir', default='./data/VOC2012/SegmentationClass', type=str)

parser.add_argument('--logfile', default='', type=str)
parser.add_argument('--comment', default='', type=str)

parser.add_argument('--mode', default='npy', type=str) # png
parser.add_argument('--max_th', default=0.50, type=float)

parser.add_argument('--bg_dir', default='', type=str)
parser.add_argument('--with_bg_cues', default=False, type=bool)
Dear author, can you explain this parameter for me?

[liaochuanlin]

Refine the background cues You can use the extracted background cues as pseudo supervision signal to train a saliency detector like PoolNet to further refine the background cues and we provide the code for background cues refinement in the directory ./PoolNet. We also provide our refined background cues at here.Does this have to run?
Will it lead to bad results if it doesn't run?

[Sierkinhane]

Could you provide some predicted results of your method and the extracted background cues?

[liaochuanlin]

/home/lcl/anaconda3/envs/BECO/bin/python3.10 /home/lcl/BECO/BECO-main/CCAM-master/WSSS/train_CCAM_VOC12.py --tag CCAM_VOC12_MOCO --pretrained mocov2 --alpha 0.25
[i] CCAM_VOC12_MOCO

[i] mean values is [0.485, 0.456, 0.406]
[i] std values is [0.229, 0.224, 0.225]
[i] The number of class is 20
[i] train_transform is Compose(
Resize(size=(512, 512), interpolation=bilinear, max_size=None, antialias=warn)
RandomHorizontalFlip(p=0.5)
RandomCrop(size=(448, 448), padding=None)
ToTensor()
Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225))
)
[i] test_transform is Compose(
<tools.ai.augment_utils.Normalize_For_Segmentation object at 0x7fe2fb7bae90>
<tools.ai.augment_utils.Top_Left_Crop_For_Segmentation object at 0x7fe2fb7bb100>
<tools.ai.augment_utils.Transpose_For_Segmentation object at 0x7fe2fb7bb340>
)
[i] #train data
[i] #valid data

[i] log_iteration : 66
[i] val_iteration : 331
[i] max_iteration : 3,310
Loading unsupervised mocov2 pretrained parameters!

[i] Architecture is resnet50
[i] Total Params: 23.54M

[i] Epoch[0/10], iteration=100, learning_rate=0.0010, loss=1.1635, positive_loss=0.0400, negative_loss=0.9580, time=40sec
[i] Epoch[0/10], iteration=200, learning_rate=0.0009, loss=1.0110, positive_loss=0.0428, negative_loss=0.9016, time=38sec
[i] Epoch[0/10], iteration=300, learning_rate=0.0009, loss=0.9442, positive_loss=0.0486, negative_loss=0.7862, time=38sec
Is Negative: True
[i] save model
[i] Epoch[1/10], iteration=100, learning_rate=0.0009, loss=0.8853, positive_loss=0.0506, negative_loss=0.7656, time=51sec
[i] Epoch[1/10], iteration=200, learning_rate=0.0009, loss=0.8497, positive_loss=0.0600, negative_loss=0.6674, time=39sec
[i] Epoch[1/10], iteration=300, learning_rate=0.0008, loss=0.8274, positive_loss=0.0543, negative_loss=0.6699, time=39sec
[i] save model
[i] Epoch[2/10], iteration=100, learning_rate=0.0008, loss=0.7983, positive_loss=0.0608, negative_loss=0.6137, time=51sec
[i] Epoch[2/10], iteration=200, learning_rate=0.0008, loss=0.7801, positive_loss=0.0557, negative_loss=0.6619, time=39sec
[i] Epoch[2/10], iteration=300, learning_rate=0.0007, loss=0.7643, positive_loss=0.0592, negative_loss=0.6105, time=39sec
[i] save model
[i] Epoch[3/10], iteration=100, learning_rate=0.0007, loss=0.7495, positive_loss=0.0554, negative_loss=0.6402, time=51sec
[i] Epoch[3/10], iteration=200, learning_rate=0.0007, loss=0.7369, positive_loss=0.0654, negative_loss=0.5693, time=39sec
[i] Epoch[3/10], iteration=300, learning_rate=0.0006, loss=0.7246, positive_loss=0.0611, negative_loss=0.5761, time=39sec
[i] save model
[i] Epoch[4/10], iteration=100, learning_rate=0.0006, loss=0.7191, positive_loss=0.0576, negative_loss=0.6141, time=50sec
[i] Epoch[4/10], iteration=200, learning_rate=0.0006, loss=0.7047, positive_loss=0.0568, negative_loss=0.6006, time=38sec
[i] Epoch[4/10], iteration=300, learning_rate=0.0005, loss=0.7014, positive_loss=0.0627, negative_loss=0.5613, time=38sec
[i] save model
[i] Epoch[5/10], iteration=100, learning_rate=0.0005, loss=0.6958, positive_loss=0.0568, negative_loss=0.5652, time=51sec
[i] Epoch[5/10], iteration=200, learning_rate=0.0005, loss=0.6861, positive_loss=0.0660, negative_loss=0.5346, time=39sec
[i] Epoch[5/10], iteration=300, learning_rate=0.0004, loss=0.6819, positive_loss=0.0610, negative_loss=0.5482, time=39sec
[i] save model
[i] Epoch[6/10], iteration=100, learning_rate=0.0004, loss=0.6764, positive_loss=0.0598, negative_loss=0.5811, time=51sec
[i] Epoch[6/10], iteration=200, learning_rate=0.0004, loss=0.6716, positive_loss=0.0674, negative_loss=0.5235, time=38sec
[i] Epoch[6/10], iteration=300, learning_rate=0.0003, loss=0.6669, positive_loss=0.0519, negative_loss=0.5649, time=38sec
[i] save model
[i] Epoch[7/10], iteration=100, learning_rate=0.0003, loss=0.6641, positive_loss=0.0515, negative_loss=0.5690, time=50sec
[i] Epoch[7/10], iteration=200, learning_rate=0.0003, loss=0.6602, positive_loss=0.0650, negative_loss=0.4885, time=38sec
[i] Epoch[7/10], iteration=300, learning_rate=0.0002, loss=0.6572, positive_loss=0.0578, negative_loss=0.5534, time=38sec
[i] save model
[i] Epoch[8/10], iteration=100, learning_rate=0.0002, loss=0.6574, positive_loss=0.0664, negative_loss=0.5179, time=50sec
[i] Epoch[8/10], iteration=200, learning_rate=0.0002, loss=0.6533, positive_loss=0.0598, negative_loss=0.5154, time=39sec
[i] Epoch[8/10], iteration=300, learning_rate=0.0001, loss=0.6497, positive_loss=0.0643, negative_loss=0.4860, time=39sec
[i] save model
[i] Epoch[9/10], iteration=100, learning_rate=0.0001, loss=0.6512, positive_loss=0.0623, negative_loss=0.5379, time=50sec
[i] Epoch[9/10], iteration=200, learning_rate=0.0001, loss=0.6496, positive_loss=0.0590, negative_loss=0.5055, time=38sec
[i] Epoch[9/10], iteration=300, learning_rate=0.0000, loss=0.6486, positive_loss=0.0569, negative_loss=0.5523, time=38sec
[i] save model
CCAM_VOC12_MOCO

Process finished with exit code 0

[Sierkinhane]

I mean you can go to the directory./experiments/images/ and show whether the predicted cues are correct.

[liaochuanlin]

Yes, the colormap.jpg picture is generated in / home/lcl/BECO/BECO-main/CCAM-master/WSSS/experiments/images/CCAM_VOC12_MOCO/train/colormaps

[Sierkinhane]

That's good. How about the generated background cues? And your cams to be refined. Could you please provide some examples?

[liaochuanlin]

Yes, the png picture is generated in / home/lcl/BECO/BECO-main/CCAM-master/WSSS/experiments/predictions/CCAM_VOC12_MOCO@train@scale=0.5,1.0,1.5,2.0@t=0.3@ccam_inference_crf=10

[liaochuanlin]

May I add your contact information?
I'm also a student.

[Sierkinhane]

python3 evaluate_using_background_cues.py --experiment_name you_experiment_name --domain train --data_dir path/to/your/data --with_bg_cues True --bg_dir path/to/your/background cues in which you_experiment_name means that the directory storing your extracted cams and bg_dir indicates the background cues.

[Sierkinhane]

If your extracted cams and background cues are correct, the evaluation results may be good.

[liaochuanlin]

Okay, I'll try again. &nbsp; 爩 ***@***.*** &nbsp;

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "Jinheng ***@***.***&gt;; 发送时间: 2023年9月5日(星期二) 下午3:39 收件人: ***@***.***&gt;; 抄送: ***@***.***&gt;; ***@***.***&gt;; 主题: Re: [CVI-SZU/CCAM] Undefined functions appear when running train_CCAM_VOC12.py？ (Issue #19) If your extracted cams and background cues are correct, the evaluation results may be good. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***&gt;