YOLOX CondInst — YOLOX 实例分割

version

demo_vis

前言

  1. 本项目是自己学习实例分割时，复现的代码. 通过自己编程，让自己对实例分割有更进一步的了解。
    若想与本人探讨有关深度学习的相关知识，欢迎通过邮件交流
  2. 后续想解决模型的部署问题(c++)
  3. 后续想继续安装其他实例分割的代码

Update

  1. waiting …

Some Ideas

  1. 在写推理的代码的时候，为了兼容eval的代码将做了很多split和cat的操作，这减慢了检测的速度，如果单纯想进行推理，可以将这部分的操作简化
  2. fp16模型存在问题，等待解决

Introduction

  1. For YOLOX, I change some codes and it will lead speed up.
  2. For CondInst, I just follow AdelaiDet and keep the same parameters as it.

Content

Quick Start

Firstly, create python environment

$ conda create -n yolox_inst python=3.7 -y

Then, clone the github of the item

$ git clone https://github.com/DDGRCF/YOLOX-CondInst.git

Then, you can adjust follow the original quick start

Instruction

Demo

I prepare the shell the demo script so that you can quickly run obb demo as:

$ cd my_exps
$ bash demo_inst.sh 0 /path/to/you
# PS: 0 is to assign the train environment to 0 gpu, you can change it by youself and /path/to/you is your demo images.

Train

I define the model default training parameters as following:

model max epoch enable_mixup enable_mosaic no aug epoch
yolox_s 24 True True 5
cls_loss_weight obj_loss_weight iou_loss_weight reg_loss_weight mask_loss_weight
1.0 1.0 5.0 1.0 5.0

Of course, this group parameters is not the best one, so you can try youself. And for the quick train, I have prepare the shell scripts, too.

$ cd my_exps
$ bash train_dota_obb.sh  0

As I set parameters above with 16 batch size per gpu (2gpu), the lresults on val dataset show as following:
waiting …

Test

I just follow original evaluation to test and eval

$ cd my_exps
$ ./eval_dota_obb.sh eval/test 0
# PS: for convenience, I set default parameters. So, eval means evaluating COCO val datasets.

Ralated Hub

GitHub

View Github