# chengdazhi/Deformable-Convolution-V2-PyTorch

Deformable ConvNets V2 (DCNv2) in PyTorch

Repository: https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch
Canonical: https://ross.abutalabs.com/products/deformable-convolution-v2-pytorch
Language: Cuda
License: MIT
License Family: permissive
Last push: 2022-11-18T21:39:26+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2821, "days_push": 1384, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1484, forks 230 (observed 2026-08-28T04:04:51.380978+00:00)

## What it is
A PyTorch implementation of Deformable Convolution V2 (DCNv2) custom CUDA operators, ported from the original MXNet implementation. It provides the deformable conv layer used to reproduce results from the Deformable ConvNets v2 paper and has been integrated into mmdetection.

## Use cases
- add deformable convolution layers to a PyTorch model
- reproduce DCNv2 results from the paper
- improve object detection accuracy on COCO with deformable convs
- build custom CUDA operators for learnable sampling offsets
- use DCNv2 with mmdetection for detection experiments

## When to choose
- you need the exact DCNv2 operator behavior described in the paper, including boundary handling fixes
- you are working with mmdetection and want the deformable conv layer it was upstreamed from
- you need efficient mini-batch deformable convolution on GPU

## When to avoid
- you use a recent PyTorch version - the repo targets PyTorch 0.4.1 and 1.0 branches and is no longer actively updated
- you just want DCN in a modern detection framework - use the official mmdetection DCN configs instead
- you need CPU-only deformable convolution - this is a CUDA implementation

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, computer-vision, image-processing
- domain: deep-learning, computer-vision, machine-learning
- platform: python, cpp
- tags: pytorch, cuda, deformable-convolution, object-detection, custom-operators, mmdetection, gpu, linux

## Member repositories
- chengdazhi/Deformable-Convolution-V2-PyTorch (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:51.380978+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:33:59.303726+00:00, confidence not recorded.
  - readme: https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch (fetched 2026-08-28T04:04:51.380978+00:00, sha 1b367d00aec7)
- Data as of 2026-08-30T08:39:29.467469+00:00.