jenjsun
commented
1 year ago Hi & thanks for your interest in our project! The current release of our code doesn't support multiple identical agents noted below (as you have 3 black mice in your example (the pre-trained model is on 1 black mouse & 1 white mouse)):
We are still working on open-sourcing the version of our code that would address multiple identical agents, but we have an internal version of our code that handles this. If you'd like us to help run the multi-agent version of our code on your videos, feel free to share your videos here! https://forms.gle/k5gwNxws8xMEEzUe6
Otherwise, we can update when the multi-agent version is open-sourced!
STAR-811
Hi! That's great work! But it doesn't work on my video. After 100 epochs, the images in the sample folder and my parameter config file are shown below. I try to use the pre-train model to finetune but it doesn't work either. What can I do to improve the keypoint discovery performance on my dataset? (train file contains 40,000 images)
dataset: custom_dataset data: data/mice workers: 4 epochs: 300 start_epoch: 0 schedule: 30 batch_size: 16 lr: 0.001 momentum: 0.9 weight_decay: 0.0001 print_freq: 10 pretrained: False gpu: 0 val_schedule: 20 curriculum: 4
checkpoint: checkpoint/custom_dataset
nkpts: 15
image_size: 256 frame_gap: 15 bounding_box: False # Only False allowed for the custom dataset
perc_weight: [100.0, 1.6, 2.3, 1.8, 2.8, 100.0]
resume: checkpoint/CalMS21/checkpoint.pth.tar evaluate: False
