Skip to content

README.md

Latest commit

 

History

History
46 lines (31 loc) · 1.46 KB

README.md

File metadata and controls

46 lines (31 loc) · 1.46 KB

DeepLabCut Training & Evaluation

Train and evaluate DeepLabCut pose estimation models using COCO-formatted annotations.

Features

  • Direct training from COCO JSON format
  • Support for multiple backbone architectures
  • Automatic best model tracking
  • NME evaluation with interocular normalization

Quick Start

# Install DeepLabCut (tested version with 11.8)
uv pip install deeplabcut==3.0.0rc8

# Train a model
python train_with_coco.py --model_type hrnet_w32 --output_dir ./dlc_model

# Evaluate model performance
python evaluate_nme.py --model_dir ./dlc_model

# Resume training from checkpoint
python train_with_coco.py --resume ./dlc_model/snapshot_epoch_50.pt

# List available architectures
python train_with_coco.py --list-models

Supported Architectures

  • HRNet: hrnet_w18, hrnet_w32, hrnet_w48 - High-resolution networks
  • ResNet: resnet_50, resnet_101, resnet_152 - Classic architectures
  • RTMPose: rtmpose_s, rtmpose_m, rtmpose_x - Real-time models

Requirements

Install DeepLabCut PyTorch following their installation guide.

Key Features

  • Direct COCO Support: Works directly with COCO JSON annotations
  • Progress Tracking: Real-time training metrics with progress bars
  • Automatic Validation: Best model saved based on validation performance
  • GPU Acceleration: Full CUDA support for efficient training