birdmandayum0131 / yolact

A simple, fully convolutional model for real-time instance segmentation.

Online Multiple Object Segmentation in Mask Coefficient Space

This repo is a implementation of my master's thesis

In my thesis, I use yolact model with some tracking algorithm(like Deep SORT) to achieve online instance segmentation.
Therefore, most of the code in this repo(to implement YOLACT) is clone from dbolya/yolact.Thanks for their contribution!

And also, DCNv2 is required to implement YOLACT++ in original repo from Daniel Bolya.
But the GPU in my hand is only one RTX 3080, and it only support CUDA 11 up ; )
Consider all supporting problem about runnning environment(versions about python, pytorch, cuda, GPU),
I use:

• GPU :RTX 3080
• Python :verion 3.8
• PyTorch :1.7.0
• CUDA :11.0
• cudnn :8.0

in this repository.
If you have similar environment problem to me, you might change to this DCN version

Last, some of the code in this repo to implement tracking algorithm be like Deep SORT is refer to ifzhang/FairMOT.
I have learned a lot from it. Thanks to their contributions too!