Issues when testing the IMC-PT-SparseGM

[gtan45]

I tried to run the experiments related to IMC-PT-SparseGM dataset by executing "experiments/ngmv2-afat/vgg16_ngmv2-afat-u_imcpt_50_stage1.yaml".

When I run initially, I got the error "FileNotFoundError: [Errno 2] No such file or directory: '/root/.cache/pygmtools/data/IMC-PT-SparseGM.tar.gz'".

Then I updated the version of the pygmtools to version "0.4.3" and re-run. Then I got the following issue "AssertionError: you should not change ROOT_DIR_IMG or ROOT_DIR_NPZ unless the data have been manually downloaded".

Suggestions on how to fix the above issue and run the experiments related to IMC-PT-SparseGM dataset will be much appreciated.

[ziao-guo]

Typically, the dataset is automatically downloaded. However, it seems that some issues are causing the dataset not to be downloaded correctly. You can provide more error information to help determine the cause of the issue.
Alternatively, you are recommended to manually download the dataset. See the homepage of IMC-PT-SparseGM here.
Download and unzip the dataset and make sure it looks like data/IMC-PT-SparseGM/annotations and data/IMC-PT-SparseGM/images.

[gtan45]

Thanks for the instructions. The dataset was downloaded as per the instructions and it was possible to run imcpt_50 dataset successfully. Thanks again.
However, when trying to run the imcpt_100 dataset the following error keep occurring. Suggestions on how to fix the error will be much appreciated.

KeyError Traceback (most recent call last)
/tmp/ipykernel_594776/3847300564.py in
17 cfg.TRAIN.CLASS if x == 'train' else cfg.EVAL.CLASS,
18 cfg.PROBLEM.TYPE)
---> 19 for x in ('train', 'test')}
20 dataloader = {x: get_dataloader(image_dataset[x], shuffle=True, fix_seed=(x == 'test'))
21 for x in ('train', 'test')}

/tmp/ipykernel_594776/3847300564.py in (.0)
17 cfg.TRAIN.CLASS if x == 'train' else cfg.EVAL.CLASS,
18 cfg.PROBLEM.TYPE)
---> 19 for x in ('train', 'test')}
20 dataloader = {x: get_dataloader(image_dataset[x], shuffle=True, fix_seed=(x == 'test'))
21 for x in ('train', 'test')}

ThinkMatch/src/dataset/data_loader.py in init(self, name, bm, length, using_all_graphs, cls, problem)
35 self.problem_type = problem
36 if problem != 'MGM3':
---> 37 self.img_num_list = self.bm.compute_img_num(self.classes)
38 else:
39 self.img_num_list = self.bm.compute_img_num(self.classes[0])

/opt/conda/lib/python3.7/site-packages/pygmtools/benchmark.py in compute_img_num(self, classes)
354 if self.name != 'SPair71k':
355 for id in data_id:
--> 356 if self.data_dict[id]['cls'] == clss:
357 cls_img_num += 1
358 num_list.append(cls_img_num)

KeyError: '70392907_4845953799'

[ziao-guo]

It seems that something goes wrong in the data preprocessing stage.

Could you please provide the version of the pygmtools package you are using and how you run the imcpt_100 experiment? That will help detect the reason for the error.

[gtan45]

The pygmtools pakcage version is pygmtools==0.3.8. Also, the imcpt_100 experiment was tried to run using "python train_eval.py --cfg experiments/gcan-afat/vgg16_gcan-afat-i_imcpt_100_stage1.yaml". Thanks.

[ziao-guo]

Thanks for the info, I found the problem.

Before you run the imcpt_100 experiment, you need to go to the data/IMC-PT-SparseGM directory and remove train.json and test.json, which are generated for imcpt_50. We will fix the bug in the later version of pygmtools.

Thanks for the issue!

[gtan45]

Thanks for the instructions and prompt reply.