MMdetection trains FCOS and CenterNet on VisDrone2019

 Configuration Environment

 

  1. Python 3.5+
  2. >=PyTorch 1.1
  3. >=CUDA 9.0
  4. NCCL 2
  5. >=GCC 4.9
  6. mmcv‘’

Download the mmdetection code

git clone https://github.com/open-mmlab/mmdetection.git

Enter this mmdetection file and prepare to compile the mmdetection file

cd mmdetection

Pack the following bags,

# mmdetection的 requirements.txt里面的文件
pip install -r requrements.py

pip install cython 

pip install albumentations>=0.3.2 imagecorruptions pycocotools six terminaltables 

Installing mmcv directly will report an error in my environment

pip install mmcv

The error is long like this, it tells you that the mmcv installation failed

      RuntimeError: Error compiling objects for extension
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
error: legacy-install-failure

× Encountered error while trying to install package.
╰─> mmcv

note: This is an issue with the package mentioned above, not pip.
hint: See above for output from the failure.

The solution is to go to the official documentation of mmcv

GitHub - open-mmlab/mmcv: OpenMMLab Computer Vision Foundation

Installation — mmcv 2.0.1 documentation

Select your cuda version and torch version, and a command statement to install with pip will be generated below. You can execute mmcv on your own terminal to install it        

Convert VisDrone to COCO format

( You need to do this step, and then change the number of categories and some parameters in the coco file below, including the mean and variance (this can be left untouched first, and then changed after running), -- then run )

training model

All object detection models are placed in "/configs"

The location of coco_detection.py "configs/_base_/datasets" or "/mmdet/configs/_base_/datasets"

Processing and conversion of individual datasets "mmdet/datasets"

The train.py used to train the model is placed in "./tools"

Run the command to start training

python tools/train.py configs/fcos/fcos_r50_caffe_fpn_gn-head_1x_coco.py

All 10 FCOS models implemented by MMdetection: https://github.com/open-mmlab/mmdetection/tree/master/configs/fcos

The model with the lowest mAP in the FCOS model: https://github.com/open-mmlab/mmdetection/blob/master/configs/fcos/fcos_r50_caffe_fpn_gn-head_1x_coco.py  ( I don’t know whether to install caffe? )

2 CenterNet models implemented by MMdetection: https://github.com/open-mmlab/mmdetection/tree/master/configs/centernet

The train.py called during training is located under ./tools

https://github.com/open-mmlab/mmdetection/blob/main/tools/train.py

 Reference:

Convert the VisDrone dataset to COCO format: convert the visdrone dataset to coco format and train on mmdetection, attach the converted json file_S5242's Blog-CSDN Blog

Use MMdetection to train your own model: (detailed tutorial) mmdetection to train your own model, test, and evaluate

 

Guess you like

Origin blog.csdn.net/Albert233333/article/details/132334949