The data processing of labelme2coco is
mainly divided into two parts: 1. Use labelme to mark samples to obtain json files; 2. Convert json files to coco datasets
Two parts of source code jump:
Mark: https://github.com/wkentaro/labelme )
to coco: https://github.com/wkentaro/labelme/tree/main/examples/instance_segmentation
json to coco dataset
-
Enter the instance_segmentation folder
Open the labelme code we downloaded before , activate the environment as labelme and go to the
labelme-main/examples/instance_segmentation folder -
Explain here, after entering the instance_segmentation folder, data_annotated is the folder where your sample data and json are stored. You can put your training data and json files in this directory. data_dataset_coo is a folder generated after running, which contains sample data and coco label dataCopy the training samples to data_annotated and modify the labels.txt file
Enter the terminal, cd to the instance_segmentation directory, and enter the following code./labelme2coco.py data_annotated data_dataset_coco --labels labels.txt
After the operation is completed, a data_dataset_coo folder is generated. At this point, the coco file of the train sample is completed, and the coco file of the val sample can repeat the above steps.
3. Put it in the specified folder
The file directory is as follows, the folder in the yellow box should be created by yourself, the red box is the copied sample data generated before, the blue box is the coco label annotations.json generated in the previous step, You need to rename the content as shown and copy it to the corresponding directory! !
If you want to do good work, you must first sharpen your tools! So far, the coco instance segmentation dataset is established! ! !4. Troubleshooting
4.1 Exit without reason in the middle of the process
. Question 1: When generating the data_dataset_coco folder of the coco dataset, it automatically exits after converting some data and the annotations.json file is not generated.Solution: Data set problem, check whether each image in the data set has a corresponding json file, and delete the images without json files.
Question 2: Every time the data set is converted to xxx (such as 477), it stops automatically without any error
Solution: Find several images before and after the 477th corresponding image, there may be image problems, just delete the wrong image
Problem 3: Unknown error
Solution: Re-run the code after deleting the generated data_dataset_coco folder