Accessing your data

First, make sure you return to the terminal/VS Code view of the Studio by clicking on the appropriate icon in the right-hand tool bar.

You will see a file explorer on the left-hand side.

Lightning Pose project structure

Data for the project named <PROJ_NAME> will be stored in a directory with the following structure:

~/Pose-app/data/<PROJ_NAME>
  ├── labeled-data/
  ├── models/
  ├── videos/
  ├── videos_infer/
  ├── videos_tmp/
  ├── CollectedData.csv
  ├── label_studio_config.xml
  ├── label_studio_metadata.yaml
  ├── label_studio_tasks.pkl
  └── model_config_<PROJ_NAME>.yaml

• labeled-data/: contains one subdirectory for each video used during labeling; each of these subdirectories contains images named img<n>.png that correspond to the labels. The subdirectory also contains images preceding and succeeding each labeled frame for training context models.

• models/: contains one subdirectory for each model trained, with the naming convention YYYY-MM-DD/HH-MM-SS. See below for more information on the files inside each model directory.

• videos/: videos uploaded for labeling. When training models, after training completion inference is automatically run on each of these videos, and the results are saved in the model-specific directory

• videos_infer/: videos uploaded for inference

• videos_tmp/: temporary directory that stores original videos before they are automatically reformatted to be compatible with DALI video readers (h.264 encoding, yuv420p pixel format). ORIGINAL VIDEOS ARE NOT SAVED.

• CollectedData.csv: hand labels in human-readable format.

• label_studio_config.xml: config info for label studio, including keypoint names and labeler instructions.

• label_studio_metadata.yaml: info on number of labeled and total tasks.

• label_studio_tasks.pkl: reformatted version of labeling tasks, also stored in label studio database.

• model_config_<PROJ_NAME>.yaml: config file for pose estimation networks; contains editable information about min/max epochs, train/val splits, network architecture, etc.

Model directory structure

Each model and its associated outputs will be stored in a directory with the following structure:

~/Pose-app/data/<PROJ_NAME>/models/YYYY-MM-DD/HH-MM-SS
  ├── tb_logs/
  ├── video_preds/
  ├── video_preds_infer/
  ├── config.yaml
  ├── predictions.csv
  ├── predictions_pca_multiview_error.csv
  ├── predictions_pca_singleview_error.csv
  └── predictions_pixel_error.csv

• tb_logs/: model weights

• video_preds/: predictions and metrics from all videos in videos directory of the project

• video_preds_infer/: predictions and metrics from all videos in videos_infer directory of the project

• config.yaml: copy of the config file used to train the model

• predictions.csv: predictions on labeled data in CollectedData.csv

• predictions_pca_multiview_error.csv: if applicable, pca multiview reprojection error for labeled data

• predictions_pca_singleview_error.csv: if applicable, pca singleview reprojection error for labeled data

• predictions_pixel_error.csv: pixel error for labeled data