voc test 报错问题

[cm0561]

voc跑测试集， 测试集真实值在官网 ，但是配置文件需要填写测试集路径，没有怎么填写呢，不填的话就会报错，很奇怪 ，求解，求指导

[mm-assistant]

We recommend using English or English & Chinese for issues so that we could have broader discussion.

[cm0561]

when wen test the test image,the cfg need to add the path of the ground truth of the test image,but we don't have the ground truth,because only the official web hava it ,if we dont add the path of the ground turth,it will present error: Not implementedError

[MeowZheng]

I have checked the config file of voc dataset, but I found the path for the test dataset.

mmsegmentation/configs/_base_/datasets/pascal_voc12.py

Lines 44 to 57 in 5c113d9

	val=dict(
	type=dataset_type,
	data_root=data_root,
	img_dir='JPEGImages',
	ann_dir='SegmentationClass',
	split='ImageSets/Segmentation/val.txt',
	pipeline=test_pipeline),
	test=dict(
	type=dataset_type,
	data_root=data_root,
	img_dir='JPEGImages',
	ann_dir='SegmentationClass',
	split='ImageSets/Segmentation/val.txt',
	pipeline=test_pipeline))

[cm0561]

I have checked the config file of voc dataset, but I found the path for the test dataset.

mmsegmentation/configs/_base_/datasets/pascal_voc12.py

Lines 44 to 57 in 5c113d9

	val=dict(
	type=dataset_type,
	data_root=data_root,
	img_dir='JPEGImages',
	ann_dir='SegmentationClass',
	split='ImageSets/Segmentation/val.txt',
	pipeline=test_pipeline),
	test=dict(
	type=dataset_type,
	data_root=data_root,
	img_dir='JPEGImages',
	ann_dir='SegmentationClass',
	split='ImageSets/Segmentation/val.txt',
	pipeline=test_pipeline))

so the problem is that, the cfg file have the path of ann_dir of test, but auctually the voc dataset don't have the ann_dir of the test, so it will present error when we perform test ofr the voc dataset.

[MeowZheng]

Might check the cfg with the below code:

from mmcv.utils import Config

cfg = Config.fromfile('cfg_file')

print(cfg.data.test.ann_dir)

Besides, based on my limited knowledge, I think Not implementedError might occur when the function or class misses, and the path of the dataset might not the main problem. Would you mind provide the whole error log?

[cm0561]

Might check the cfg with the below code:

from mmcv.utils import Config

cfg = Config.fromfile('cfg_file')

print(cfg.data.test.ann_dir)

Besides, based on my limited knowledge, I think Not implementedError might occur when the function or class misses, and the path of the dataset might not the main problem. Would you mind provide the whole error log?

The whole error code is below:
warnings.warn( [ ] 0/1456, elapsed: 0s, ETA:/home/clz/.local/lib/python3.8/site-packages/torch/nn/functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at /pytorch/c10/core/TensorImpl.h:1156.) return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode) Traceback (most recent call last): File "mmsegmentation/tools/test.py", line 320, in <module> main() File "mmsegmentation/tools/test.py", line 273, in main results = single_gpu_test( File "/home/clz/mmsegmentation/mmseg/apis/test.py", line 123, in single_gpu_test result = dataset.format_results( File "/home/clz/mmsegmentation/mmseg/datasets/custom.py", line 252, in format_results raise NotImplementedError

[MeowZheng]

mmsegmentation/mmseg/datasets/custom.py

Lines 250 to 252 in 5c113d9

	def format_results(self, results, imgfile_prefix, indices=None, **kwargs):
	"""Place holder to format result to dataset specific output."""
	raise NotImplementedError

[cm0561]

So,how can i solve this problem

[MeowZheng]

might check this arg

mmsegmentation/mmseg/apis/test.py

Line 41 in 5c113d9

format_only=False,

or just remove these lines

mmsegmentation/mmseg/apis/test.py

Lines 122 to 124 in 5c113d9

	if format_only:
	result = dataset.format_results(
	result, indices=batch_indices, **format_args)

[cm0561]

might check this arg

mmsegmentation/mmseg/apis/test.py

Line 41 in 5c113d9

format_only=False,

or just remove these lines

mmsegmentation/mmseg/apis/test.py

Lines 122 to 124 in 5c113d9

	if format_only:
	result = dataset.format_results(
	result, indices=batch_indices, **format_args)

ok thanks ,I will try right now, I think there is something wrong with the code of voc dataset,can the official fix it!

[cm0561]

might check this arg

mmsegmentation/mmseg/apis/test.py

Line 41 in 5c113d9

format_only=False,

or just remove these lines

mmsegmentation/mmseg/apis/test.py

Lines 122 to 124 in 5c113d9

	if format_only:
	result = dataset.format_results(
	result, indices=batch_indices, **format_args)

ok thanks ,I will try right now, I think there is something wrong with the code of voc dataset,can the official fix it!

I just try the method that remove the three lines, but this function is to save the seg_map,I remove It, the error disappear,but the seg_map disappear too. but my purpose is to have the seg_map

[MeowZheng]

VOC dataset didn't support saving the predicted results, and we will support it asap, or if you have interests, could you please create a nice pr for this function, which only needs to implement format_results in VOC dataset.