DOCSEARCHbyAlgolia
Submit Docs

ptlflow

repository·main·Indexed 19 days ago

https://github.com/hmorimitsu/ptlflow

A unified PyTorch Lightning-based framework for training and evaluating various state-of-the-art optical flow models. Version 0.4.2 includes implementations of models such as CCMR, CRAFT, CSFlow, DICL-Flow, DIP, DPFlow, FastFlowNet, FlowFormer, FlowFormer++, FlowSeek, GMA, GMFlow, GMFlowNet, HD3, IRR, LCV + RAFT, LiteFlowNet (1, 2, and 3), and MaskFlownet.

Tokens
20.8K
Snippets
65
Records
137
Agent score
67%

What's inside ptlflow

  1. Overview of LiteFlowNet models

    main

    The ptlflow repository provides implementations for the LiteFlowNet family of lightweight convolutional neural networks for optical flow estimation. This includes three distinct versions:

    • LiteFlowNet: The original lightweight CNN for optical flow.
    • LiteFlowNet2: A version revisiting data fidelity and regularization.
    • LiteFlowNet3: A version focused on resolving correspondence ambiguity for improved accuracy.

    Users can refer to the original research repositories for the source code of these models.

  5. Overview of PyTorch Lightning Optical Flow (ptlflow)

    main
    ptlflow is a unified framework designed for training and testing a collection of state-of-the-art deep learning models for optical flow estimation. It leverages PyTorch Lightning to provide a consistent interface across many different models and datasets, making it easier for researchers to compare and deploy various optical flow algorithms.

  11. Use the IRR model for Optical Flow and Occlusion Estimation

    main
    The IRR (Iterative Residual Refinement) model is available in ptlflow for joint optical flow and occlusion estimation. This implementation is based on the original research: Iterative Residual Refinement for Joint Optical Flow and Occlusion Estimation (CVPR 2019).