Skip to content

Latest commit

 

History

History
79 lines (65 loc) · 5.74 KB

pose_tracking_gui_tutorial.md

File metadata and controls

79 lines (65 loc) · 5.74 KB

Pose tracking (GUI) 🐭

tracker

The latest python version is integrated with Facemap network for tracking 14 distinct keypoints on mouse face and an additional point for tracking paw. The keypoints can be tracked from different camera views (some examples shown below).

view1 view2

Generate keypoints

Follow the steps below to generate keypoints for your videos or download sample video from here:

  1. Load video

    • Select File from the menu bar
    • For processing single video, select Load video. Alternatively, for processing multiple videos, select Load multiple videos to select the folder containing the videos. (Note: Pose estimation for multipl videos is only supported for videos recorded simultaneously i.e. have the same time duration and frame rate).

    (Optional) Set output folder

    • Use the file menu to Set output folder.
    • The processed keypoints (*.h5) and metadata (*.pkl) will be saved in the selected output folder or folder containing the video (default).

  2. Process video(s)

    • Check Keypoints for pose tracking.
    • Click process.

  3. Set ROI/bounding box for face region

    • A dialog box for selecting a bounding box for the face will appear. Drag the red rectangle to select region of interest on the frame where the keypoints will be tracked. Please ensure that the bouding box is focused on the face where all the keypoints will be visible. See example frames here. If a 'Face (pose)' ROI has already been added then this step will be skipped.
    • Click Done to process video. Alternatively, click Skip to use the entire frame region. Monitor progress bar at the bottom of the window for updates.

  4. View keypoints

    • Keypoints will be automatically loaded after processing.
    • Processed keypoints file will be saved as [videoname]_FacemapPose.h5 in the selected output folder.

Visualize keypoints

To load keypoints (*.h5) for a video generated using Facemap or other software in the same format (such as DeepLabCut and SLEAP), follow the steps below:

  1. Load video
    • Select File from the menu bar
    • Select Load video
  2. Load keypoints
    • Select Pose from the menu bar
    • Select Load keypoints
    • Select the keypoints (*.h5) file
  3. View keypoints
    • Use the "Keypoints" checkbox to toggle the visibility of keypoints.
    • Change value of "Threshold (%)" under pose settings to filter keypoints with lower confidence estimates. Higher threshold will show keypoints with higher confidence estimates.

Note: this feature is currently only supported for single video. Please see CLI instructions for viewing keypoints for multiple videos.

Finetune model to refine keypoints for a video

To improve keypoints predictions for a video, follow the steps below:

  1. Load video
    • Select File from the menu bar
    • Select Load video
  2. Set finetuned model's output folder
    • Select Pose from the menu bar
    • Select Finetune model
    • Set output folder path for finetuned model
  3. Select training data and set training parameters
    • Set Initial model to use for training. By default, Facemap's base model trained on our dataset will be used for fine-tuning. Alternatively, you can select a model previously finetuned on your own dataset.
    • Set Output model name for the finetuned model.
    • Choose Yes/No to refine keypoints prediction for the video loaded and set # Frames to use for training. You can also choose proportion of random vs. outlier frames to use for training. The outlier frames are selected using the Difficulty threshold (percentile), which determines the percentile of confidence scores to use as the threshold for selecting frames with the highest error.
    • Choose Yes/No to add previously refined keypoints to the training set.
    • Set training parameters or use default values.
    • Click Next
  4. Refine keypoints
    • If a ROI/bounding box was not added, then a dialog box for selecting a bounding box for the face will appear. Drag the red rectangle to select region of interest on the frame where the keypoints will be tracked.
    • Click Done to process video. Alternatively, click Skip to use the entire frame region. Monitor progress bar at the bottom of the window for updates.
    • Drag keypoints to refine predictions. Use Shift+D to delete a keypoint. Right click to add a deleted keypoint. Use Previous and Next buttons to change frame. Click Help for more details.
    • Click Train model to start training. A progress bar will appear for training updates.
  5. Evaluate training
    • View predicted keypoints for test frames from the video loaded. For further refinement, Click Continue training that will repeat steps 3-5.
    • Click Save model to save the finetuned model. The finetuned model will be saved as *.pt in the selected output folder.
  6. Generate keypoints using the finetuned model
    • Use the Pose model dropdown menu to set the finetuned model to use for generating keypoints predictions.
    • (Optional) Change "Batch size" under pose settings.
    • Click Process to generate keypoints predictions. See Generate keypoints for more details.