samchung95 / til-23-cv

TIL2023 CV Qualifiers Challenge

The TIL2023 CV Qualifiers code repository

Setting up the environment

Clone Repository

Run this command to clone the repository

git clone https://github.com/til-23/til-23-cv.git

Install requirements

To install the requirements, create a virtual environment and install the yolov5 requirements. To use our pretrained weights, you also need to install yolov5.

Model Training

Object detection

We provide yolov5 pretrained weights for you to finetune your models as a base, but you are free to use other object detection libraries. To finetune the weights on your dataset, run the following command from the yolov5 repo.

python train.py --data coco.yaml --epochs 300 --weights 'pretrained_weights.pt' --cfg yolov5n.yaml  --batch-size 128

You can also refer to this tutorial on training a yolov5 model.

Object Re-Identification

Refer to src/reID. The directory contains the following files:

• dataset.py - This file converts your images into a torch.utils.data.Dataset class. You will need to have your cropped images of your plushies and in the LFW format for it to be compatible.
• transforms.py - This file preprocesses your images to ensure they're ingestible by the model. The most important preprocessing step is to resize the image to a standard size before they're passed into the model.
• model.py - This file contains the Siamese Network. This is the model you will train.
• train.py - This file contains the code to fit your model to the dataset.
• test.py - This file lets you test your model on a pair of plushie images.
• utils.py - This file contains misc functions that you could use.
• model.pth - A pretrained reID model as a baseline

Model Inference

We have created a boilerplate code that allows you to detect plushies in a scene, and ReID a particular plushie from the detected plushies:

python3 src/inference.py