SEMANTIC SEGMENTATION

Live workings of the semantic segmentation model.

semseg.mp4

drivesemseg.mp4

Prerequisites

THANK, WORSHIP, PRAY to Tamoghna!!!

This requires Python 3.7+, CUDA 10.2+ and PyTorch 1.8+.

Installing pytorch

pip install torch;
pip install torchvision

Installing dependencies :

pip install testresources ;
pip install launchpadlib ;
pip install --upgrade pip setuptools ;
pip install --upgrade six 

Installation

Step 1. Installing mmengine and mmcv using openmim

pip install -U openmim ;
mim install mmengine ;
mim install "mmcv>=2.0.0" 

Step 2. Install MMSegmentation.

cd semseg ; 
pip install -v -e .
# '-v' means verbose, or more output
# '-e' means installing a project in editable mode,
# thus any local modifications made to the code will take effect without reinstallation.

Verify the installation

To verify whether MMSegmentation is installed correctly, we provide some sample codes to run an inference demo.

Step 0. Checking mmseg version.

import mmseg
print(mmseg.__version__)
# Example output: 1.0.0

The following steps to verify installation are optional

Step 1. We need to download config and checkpoint files.

mim download mmsegmentation --config pspnet_r50-d8_4xb2-40k_cityscapes-512x1024 --dest .

The downloading will take several seconds or more, depending on your network environment. When it is done, you will find two files pspnet_r50-d8_4xb2-40k_cityscapes-512x1024.py and pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth in your current folder.

Step 2. Verify the inference demo.

Option (a). If you install mmsegmentation from source, just run the following command.

python demo/image_demo.py demo/demo.png configs/pspnet/pspnet_r50-d8_4xb2-40k_cityscapes-512x1024.py pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth --device cuda:0 --out-file result.jpg

You will see a new image result.jpg on your current folder, where segmentation masks are covered on all objects.

Running semantic segmentation.

Run semseg.py

Plausible semseg models :

Option(A). configs/bisenetv1/bisenetv1_r18-d32-in1k-pre_4xb4-160k_cityscapes-1024x1024.py && bisenetv1_r18-d32_in1k-pre_4x4_1024x1024_160k_cityscapes_20210905_220251-8ba80eff.pth :
gives a decent computational speed of semseg although the accuracy is compromised a bit (70%).

Option(B). configs/pspnet/pspnet_r50b-d8_4xb2-80k_cityscapes-512x1024.py && pspnet_r50b-d8_512x1024_80k_cityscapes_20201225_094315-6344287a.pth :
this one is found to have the highest amount accuracy(81%) so far but very very slow computational speed something around 3 fps which is very bad.

Option(C). configs/ddrnet/ddrnet_23-slim_in1k-pre_2xb6-120k_cityscapes-1024x1024.py && ddrnet_23-slim_in1k-pre_2xb6-120k_cityscapes-1024x1024_20230426_145312-6a5e5174.pth :
model link : https://download.openmmlab.com/mmsegmentation/v0.5/ddrnet/ddrnet_23-slim_in1k-pre_2xb6-120k_cityscapes-1024x1024/ddrnet_23-slim_in1k-pre_2xb6-120k_cityscapes-1024x1024_20230426_145312-6a5e5174.pth works with moderate accuracy and moderate speed. I will be using this for now. DISCLAIMER : Need the bolt zed cam param tuning for this to work well!!!