How to build a semi-supervised data set

[1999luodi]

Where does the unlabeled2017 data of semi-supervised data set come from? Is it independent of the train data

At the beginning coco’s dataset:
mmdetection
├── data
│ ├── coco
│ │ ├── annotations
│ │ │ ├── image_info_unlabeled2017.json
│ │ │ ├── instances_train2017.json
│ │ │ ├── instances_val2017.json
│ │ ├── test2017
│ │ ├── train2017
│ │ ├── unlabeled2017
│ │ ├── val2017

After processing the semi-supervised json:
mmdetection
├── data
│ ├── coco
│ │ ├── annotations
│ │ │ ├── image_info_unlabeled2017.json
│ │ │ ├── instances_train2017.json
│ │ │ ├── instances_val2017.json
│ │ ├── semi_anns
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ │ ├── [email protected]
│ │ ├── test2017
│ │ ├── train2017
│ │ ├── unlabeled2017
│ │ ├── val2017

The data set used at the end：
mmdetection
├── data
│ ├── coco
│ │ ├── annotations
│ │ │ ├── image_info_unlabeled2017.json
│ │ │ ├── instances_train2017.json
│ │ │ ├── instances_unlabeled2017.json
│ │ │ ├── instances_val2017.json
│ │ ├── test2017
│ │ ├── train2017
│ │ ├── unlabeled2017
│ │ ├── val2017

among unlabeled2017，image_info_unlabeled2017.json，instances_unlabeled2017.json，how these file make? And why the semi_anns file disappear？

[1999luodi]

when I run train.py，I get an error where the program breaks at 150 iterations or 300 iterations with an error that it can't find loss_cla. Then I use tools/analysis_tools/browse_dataset.py and it also reports the error: warnings.warn(f'Failed to add {vis_backend.class}, ').
[0/6305, elapsed: 0s, ETA: Traceback (most recent call last):
File “tools/analysis_tools/browse_dataset.py”, line 89, in
main()
File “tools/analysis_tools/browse_dataset.py”, line 57, in main
img = item['inputs'].permute(1, 2, 0).numpy()
AttributeError: 'dict' object does not have 'permute' attribute

[1999luodi]

大佬，求帮助

[1999luodi]

@Keiku @jbwang1997

[1999luodi]

god，help please!

[1999luodi]

I also used the raw dataset (not converted to semi-supervised format) for browse_dataset.py It works, but with some warnings: UserWarning: Warning: polygon out of bounds, plotted polygons may not be in the image and UserWarning: Warning: box out of bounds, plotted boxes may not be in the image