OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Developer Tools & SDKs

Mmdeploy

Mmdeploy is an open-source Python framework for converting and deploying supported OpenMMLab computer-vision models with inference backends such as ONNX Runtime, OpenVINO, ncnn, PPLNN, LibTorch and TVM.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Mmdeploy — OpenViST project profile in AI Developer Tools & SDKs.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Mmdeploy when a supported OpenMMLab model needs conversion or deployment through one of the listed inference backends. Check the supported-codebase documentation and platform matrix before selecting a target.

Who it is for

  • Developers building deployment workflows for supported OpenMMLab models.
  • Teams selecting an inference backend for computer-vision models.

Documented uses

  • Convert a supported OpenMMLab model for an inference backend.
  • Build a deployment pipeline around supported models and backends.
  • Check model and backend combinations against the project's device-platform matrix.
Before you choose

Requirements & limits

The repository is public and lists an Apache-2.0 license.

  • Check that the model's OpenMMLab codebase appears in the supported list.
  • Backend support depends on the device and platform combination; consult the repository's matrix.
  • Align MMDeploy versions with the relevant OpenMMLab dependencies: the README distinguishes 1.x and 0.x compatibility lines.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up MMDeploy for Linux x86_64 CPU inference with ONNX Runtime

Install the documented MMDeploy converter and ONNX Runtime SDK packages for a Linux x86_64 CPU setup.

Before you start

  • Python 3.6 or later and PyTorch 1.8 or later for end-to-end deployment.
  • A Linux x86_64 CPU environment.
  • Miniconda, MMCV, and ONNX Runtime.

Steps

  1. Install Miniconda, create and activate a Python 3.8 conda environment, and install PyTorch following its official instructions; for CPU platforms, the guide gives conda install pytorch=={pytorch_version} torchvision=={torchvision_version} cpuonly -c pytorch.

  2. Install MMCV prerequisites with the command below, mim install mmengine, and mim install "mmcv>=2.0.0rc2".

    Documented command
    pip install -U openmim

  3. Install the converter and CPU ONNX Runtime SDK packages with the command below and the command below.

    Documented command
    pip install mmdeploy==1.3.1
    Documented command
    pip install mmdeploy-runtime==1.3.1

  4. Install ONNX Runtime 1.8.1 by downloading and extracting its Linux x64 archive, then set ONNXRUNTIME_DIR to the extracted directory and add its lib directory to LD_LIBRARY_PATH.

Expected outcome

The documented Linux x86_64 CPU package setup for the MMDeploy converter and ONNX Runtime SDK is in place.

Guide limitations

  • The package commands are given as a latest-precompiled-package example using MMDeploy 1.3.1; align MMDeploy and OpenMMLab component versions.
  • The ONNX Runtime archive instructions specify version 1.8.1.
  • This recipe documents package setup only, not model conversion or inference.
Documentation 1 ↗
Source-based practical guide

Convert a Faster R-CNN model to TensorRT with MMDeploy

Convert the documented MMDetection Faster R-CNN checkpoint into a TensorRT deployment model.

Before you start

  • A configured Linux x86_64 CUDA environment with MMDeploy 1.3.1, MMCV, and the required OpenMMLab components installed.
  • TensorRT 8.2.3.0, CUDA 11.x, cuDNN 8.2.1, and PyCUDA, as specified in the guide.
  • The Faster R-CNN model configuration, checkpoint, demo image, and MMDeploy deployment configuration used in the steps.

Steps

  1. Clone the MMDeploy main branch and MMDetection 3.x branch; from the MMDetection directory, run mim install -v -e ..

  2. Download the Faster R-CNN checkpoint to a checkpoints directory using the checkpoint URL in the documentation.

  3. From the parent directory, run `python mmdeploy/tools/deploy.py mmdeploy/configs/mmdet/detection/detection_tensorrt_dynamic-320x320-1344x1344.py mmdetection/configs/faster_rcnn/faster-rcnn_r50_fpn_1x_coco.py checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth mmdetection/demo/demo.jpg --work-dir mmdeploy_mode

  4. Use the generated deployment model and metadata in the --work-dir path as the input to the documented MMDeploy SDK inference stage.

Expected outcome

The converted TensorRT backend model and its MMDeploy metadata are placed under `mmdeploy_model/faster-rcnn`.

Guide limitations

  • The example depends on the specified MMDetection model and deployment configuration and a CUDA/TensorRT environment.
  • TensorRT conversion requires the documented TensorRT, CUDA, cuDNN, and PyCUDA setup.
  • The guide says the generated model directory, rather than only the engine file, is used as the SDK model path; inference itself is not included in this recipe.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
MmdeployConvert a supported OpenMMLab model for an inference backend.Linux, macOS, Self-hosted, ServerThe repository is public and lists an Apache-2.0 license.
HyperposeCustomize pose-estimation architectures and datasets.Docker, Linux, macOS, Windows
Multi Task NlpDefine multiple tasks and their data and model settings in a YAML file.Linux, macOS, WindowsThe project is an open-source library under the Apache-2.0 license, distributed through its public GitHub repository.
Evidence and transparency

Sources & review

Official repository ↗Guide documentation ↗
Source record details
Source availability
Active
Sources checked
15 October 2026
Original website title
open-mmlab/mmdeploy · Official homepage title · retained in its source language
How we review sourcesSuggest a correction