about run run-wfdd.sh

[Kingxudong]

I ran your code, the data was downloaded, but the following error occurred:

0%| | 0/640 [00:59<?, ?epoch/s] Traceback (most recent call last): File "F:\SICC\GLASS-main\GLASS-main\main.py", line 351, in main() File "D:\annaconda\lib\site-packages\click\core.py", line 1157, in call return self.main(args, kwargs) File "D:\annaconda\lib\site-packages\click\core.py", line 1078, in main rv = self.invoke(ctx) File "D:\annaconda\lib\site-packages\click\core.py", line 1720, in invoke return _process_result(rv) File "D:\annaconda\lib\site-packages\click\core.py", line 1657, in _process_result value = ctx.invoke(self._result_callback, value, ctx.params) File "D:\annaconda\lib\site-packages\click\core.py", line 783, in invoke return __callback(args, **kwargs) File "F:\SICC\GLASS-main\GLASS-main\main.py", line 299, in run flag = GLASS.trainer(dataloaders["training"], dataloaders["testing"], dataset_name) File "F:\SICC\GLASS-main\GLASS-main\glass.py", line 255, in trainer for i, data in enumerate(training_data): File "D:\annaconda\lib\site-packages\torch\utils\data\dataloader.py", line 631, in next data = self._next_data() File "D:\annaconda\lib\site-packages\torch\utils\data\dataloader.py", line 1346, in _next_data return self._process_data(data) File "D:\annaconda\lib\site-packages\torch\utils\data\dataloader.py", line 1372, in _process_data data.reraise() File "D:\annaconda\lib\site-packages\torch_utils.py", line 722, in reraise raise exception FileNotFoundError: Caught FileNotFoundError in DataLoader worker process 0. Original Traceback (most recent call last): File "D:\annaconda\lib\site-packages\torch\utils\data_utils\worker.py", line 308, in _worker_loop data = fetcher.fetch(index) File "D:\annaconda\lib\site-packages\torch\utils\data_utils\fetch.py", line 51, in fetch data = [self.dataset[idx] for idx in possibly_batched_index] File "D:\annaconda\lib\site-packages\torch\utils\data_utils\fetch.py", line 51, in data = [self.dataset[idx] for idx in possibly_batched_index] File "F:\SICC\GLASS-main\GLASS-main\datasets\mvtec.py", line 192, in getitem mask_fg = PIL.Image.open(fgmask_path) File "D:\annaconda\lib\site-packages\PIL\Image.py", line 3236, in open fp = builtins.open(filename, "rb") FileNotFoundError: [Errno 2] No such file or directory: 'F:/SICC/GLASS-main/GLASS-main/datasets/WFDD\fg_mask/grey_cloth/129.png'

[cqylunlun]

According to the error message, you need to download the Foreground Mask (Download link), as described in the 4^th section of Dataset Release. After the download is complete, please move the ./All_fg_mask/WFDD/fg_mask directory from the downloaded folder to your designated WFDD path, resulting in F:/SICC/GLASS-main/GLASS-main/datasets/WFDD/fg_mask.

If you prefer not to use the fg_mask, you can simply set the argument --fg to ’0‘ in the run-wfdd.sh script.

[Kingxudong]

"Thank you very much for your reply. Through training, I have seen excellent metric results, which is outstanding work. However, when performing individual tests, the results are not what I expected. Could you please share the test code? Or how should I configure it?"？

[cqylunlun]

Thank you for your recognition. For individual tests, you can change the argument --test to 'test' in run-wfdd.sh. Normally, the test results should be consistent with the training results. If you encounter any test issues, please check and debug method 'tester' in glass.py, which contains the test code.

[Kingxudong]

Thank you very much. I have saved norm_segmentations in the _evaluate function in glass.py, but it seems that the results are not correct.

def _evaluate(self, images, scores, segmentations, labels_gt, masks_gt, name, path='training'): scores = np.squeeze(np.array(scores)) img_min_scores = min(scores) img_max_scores = max(scores) norm_scores = (scores - img_min_scores) / (img_max_scores - img_min_scores + 1e-10)

    image_scores = metrics.compute_imagewise_retrieval_metrics(norm_scores, labels_gt, path)
    image_auroc = image_scores["auroc"]
    image_ap = image_scores["ap"]

    if len(masks_gt) > 0:

        segmentations = np.array(segmentations)
        min_scores = np.min(segmentations)
        max_scores = np.max(segmentations)
        norm_segmentations = (segmentations - min_scores) / (max_scores - min_scores + 1e-10)

        for i, seg in enumerate(norm_segmentations):

            img_up = cv2.resize(seg, (256, 256))

            output_path = './results/segmentations/{}.png'.format(i + 1)
            cv2.imwrite(output_path, (img_up * 255).astype(np.uint8))  # Scale back to [0, 255] for saving
        pixel_scores = metrics.compute_pixelwise_retrieval_metrics(norm_segmentations, masks_gt, path)
        pixel_auroc = pixel_scores["auroc"]
        pixel_ap = pixel_scores["ap"]
        if path == 'eval':
            try:
                pixel_pro = metrics.compute_pro(np.squeeze(np.array(masks_gt)), norm_segmentations)

            except:
                pixel_pro = 0.
        else:
            pixel_pro = 0.

    else:
        pixel_auroc = -1.
        pixel_ap = -1.
        pixel_pro = -1.
        return image_auroc, image_ap, pixel_auroc, pixel_ap, pixel_pro

    defects = np.array(images)
    targets = np.array(masks_gt)
    for i in range(len(defects)):
        defect = utils.torch_format_2_numpy_img(defects[i])
        target = utils.torch_format_2_numpy_img(targets[i])

        mask = cv2.cvtColor(cv2.resize(norm_segmentations[i], (defect.shape[1], defect.shape[0])),
                            cv2.COLOR_GRAY2BGR)
        mask = (mask * 255).astype('uint8')

        mask = cv2.applyColorMap(mask, cv2.COLORMAP_JET)

        img_up = np.hstack([defect, target, mask])
        img_up = cv2.resize(img_up, (256 * 3, 256))
        full_path = './results/' + path + '/' + name + '/'
        utils.del_remake_dir(full_path, del_flag=False)
        cv2.imwrite(full_path + str(i + 1).zfill(3) + '.png', img_up)

    return image_auroc, image_ap, pixel_auroc, pixel_ap, pixel_pro

[cqylunlun]

norm_segmentations is used to calculate metrics and output segmentation heatmaps. The best metrics in both training and testing are obtained using the same model through method _evaluate. Theoretically, the metric and heatmap results should be the same. Could you explain your issue again in more detail?

[Kingxudong]

I need to save the binary images of the test results. I tried saving segmentations, but it’s not the segmentation result. How should I save the binary images after segmentation?

[cqylunlun]

After the test is completed, the results of the original image/ground truth/heatmap have been saved in ./results/eval/wfdd_*. If you need binary segmentation maps instead of heatmap comparison images, you can modify lines L530~L543 of glass.py, specifically the conversion code from norm_segmentations[i] to mask in lines L534~L537.