how to deal with instance with 3 points in MMOCR 1.x det #1595

jorie-peng · 2022-12-05T10:02:10Z

jorie-peng
Dec 5, 2022

dataset: total-text
env: MMOCR dev-1.x
model: fcenet
config: as same as fcenet_fpn_ic15
error: because total-text has instance with 1/2/3 points,

step1. if train fcenet with dev-1.x totaltext dataset directly, it will have error , it seems must have points > 6。

TopologyException: Input geom 0 is invalid: Self-intersection at 95.599836545279544 343.73606162303474
Traceback (most recent call last):
File "tools/train.py", line 117, in
main()
File "tools/train.py", line 113, in main
runner.train()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/runner/runner.py", line 1631, in train
model = self.train_loop.run() # type: ignore
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/runner/loops.py", line 88, in run
self.run_epoch()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/runner/loops.py", line 103, in run_epoch
for idx, data_batch in enumerate(self.dataloader):
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 435, in next
data = self._next_data()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1065, in _next_data
return self._process_data(data)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1111, in _process_data
data.reraise()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/_utils.py", line 428, in reraise
raise self.exc_type(msg)
AssertionError: Caught AssertionError in DataLoader worker process 1.
Original Traceback (most recent call last):
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 198, in _worker_loop
data = fetcher.fetch(index)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/dataset/base_dataset.py", line 408, in getitem
data = self.prepare_data(idx)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/dataset/base_dataset.py", line 789, in prepare_data
return self.pipeline(data_info)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/dataset/base_dataset.py", line 58, in call
data = t(data)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmcv/transforms/base.py", line 12, in call
return self.transform(results)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmcv/transforms/processing.py", line 1551, in transform
results = self.resize(results)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmcv/transforms/base.py", line 12, in call
return self.transform(results)
File "/mnt/lustre/pengjiahui/download_dataset/OCR/mmocr/mmocr/datasets/transforms/ocr_transforms.py", line 604, in transform
self._resize_polygons(results)
File "/mnt/lustre/pengjiahui/download_dataset/OCR/mmocr/mmocr/datasets/transforms/ocr_transforms.py", line 584, in _resize_polygons
polygon = crop_polygon(polygon, crop_bbox)
File "/mnt/lustre/pengjiahui/download_dataset/OCR/mmocr/mmocr/utils/polygon_utils.py", line 153, in crop_polygon
poly = poly2shapely(polygon)
File "/mnt/lustre/pengjiahui/download_dataset/OCR/mmocr/mmocr/utils/polygon_utils.py", line 111, in poly2shapely
assert polygon.size % 2 == 0 and polygon.size >= 6
AssertionError

step2: refer to #768, change instance with points<8 to ignore=True, add filter code in /mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/dataset/base_dataset.py filter_data(), but still fail.

TopologyException: Input geom 0 is invalid: Self-intersection at 143.39978858616797 515.82527798853891
Traceback (most recent call last):
File "tools/train.py", line 117, in
main()
File "tools/train.py", line 113, in main
runner.train()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/runner/runner.py", line 1631, in train
model = self.train_loop.run() # type: ignore
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/runner/loops.py", line 88, in run
self.run_epoch()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/runner/loops.py", line 103, in run_epoch
for idx, data_batch in enumerate(self.dataloader):
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 435, in next
data = self._next_data()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1065, in _next_data
return self._process_data(data)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1111, in _process_data
data.reraise()
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/_utils.py", line 428, in reraise
raise self.exc_type(msg)
shapely.errors.TopologicalError: Caught TopologicalError in DataLoader worker process 3.
Original Traceback (most recent call last):
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 198, in _worker_loop
data = fetcher.fetch(index)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/dataset/base_dataset.py", line 413, in getitem
data = self.prepare_data(idx)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/dataset/base_dataset.py", line 794, in prepare_data
return self.pipeline(data_info)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmengine/dataset/base_dataset.py", line 58, in call
data = t(data)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmcv/transforms/base.py", line 12, in call
return self.transform(results)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmcv/transforms/processing.py", line 1551, in transform
results = self.resize(results)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/mmcv/transforms/base.py", line 12, in call
return self.transform(results)
File "/mnt/lustre/pengjiahui/download_dataset/OCR/mmocr/mmocr/datasets/transforms/ocr_transforms.py", line 604, in transform
self._resize_polygons(results)
File "/mnt/lustre/pengjiahui/download_dataset/OCR/mmocr/mmocr/datasets/transforms/ocr_transforms.py", line 584, in _resize_polygons
polygon = crop_polygon(polygon, crop_bbox)
File "/mnt/lustre/pengjiahui/download_dataset/OCR/mmocr/mmocr/utils/polygon_utils.py", line 155, in crop_polygon
poly_cropped = poly.intersection(crop_poly)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/shapely/geometry/base.py", line 695, in intersection
return geom_factory(self.impl['intersection'](self, other))
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/shapely/topology.py", line 73, in call
self._check_topology(err, this, other)
File "/mnt/cache/pengjiahui/anaconda3/envs/mmde3/lib/python3.8/site-packages/shapely/topology.py", line 38, in _check_topology
raise TopologicalError(
shapely.errors.TopologicalError: The operation 'GEOSIntersection_r' could not be performed. Likely cause is invalidity of the geometry <shapely.geometry.polygon.Polygon object at 0x7f13b8ee6700>

step3: for TopologicalError error, refer to #1388, change pipeline, but fail still, error message keep same as step2.

so how to change my config or dataset, can train successfully?

Harold-lkk · 2022-12-07T06:28:58Z

Harold-lkk
Dec 7, 2022
Collaborator

Thanks for your report! Can you share the code you change the filter_data?

4 replies

jorie-peng Dec 7, 2022
Author

filter_data code as follows:

def filter_data(self) -> List[dict]:
    for data_info in self.data_list:
        for j,instance_info in enumerate(data_info['instances']):
             polygon = instance_info['polygon']
             if len(polygon)<8:
                 del data_info['instances'][j]
    return self.data_list

Harold-lkk Dec 7, 2022
Collaborator

Thanks for your reply! We will try it and reply ASAP.

gaotongxiao Dec 28, 2022
Maintainer

We have thoroughly fixed this issue, and also released an example config for totaltext:

https://github.com/open-mmlab/mmocr/blob/dev-1.x/configs/textdet/fcenet/fcenet_resnet50_fpn_1500e_totaltext.py

jorie-peng Jan 4, 2023
Author

Thanks! I will try it!

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

how to deal with instance with 3 points in MMOCR 1.x det #1595

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Replies: 1 comment 4 replies

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{title}}

Select a reply

how to deal with instance with 3 points in MMOCR 1.x det #1595

jorie-peng Dec 5, 2022

Replies: 1 comment · 4 replies

Harold-lkk Dec 7, 2022 Collaborator

jorie-peng Dec 7, 2022 Author

Harold-lkk Dec 7, 2022 Collaborator

gaotongxiao Dec 28, 2022 Maintainer

jorie-peng Jan 4, 2023 Author

jorie-peng
Dec 5, 2022

Replies: 1 comment 4 replies

Harold-lkk
Dec 7, 2022
Collaborator

jorie-peng Dec 7, 2022
Author

Harold-lkk Dec 7, 2022
Collaborator

gaotongxiao Dec 28, 2022
Maintainer

jorie-peng Jan 4, 2023
Author