abbylagar / objectdetection

Drinks Object Detection using Faster Rcnn Resnet50 FPN

created by : Abegail Lagar

Tools used:

Pytorch Lightning
Wandb

Folder Structure

.
├── weights                   # Pretrained weights
├── outputs                   # Annotated output
├── samples                   # Raw video
├── datasets                  # Dataset folder
│   └── python          
│         └── drinks          # Drinks Dataset
├── various codes             #train ,test codes and other utilities 
└── README.md

Set-up procedures:

steps before training and testing the model

1. Clone the repository:

!git clone https://github.com/abbylagar/objectdetection.git

2. Go to your project directory

%cd <project_folder>

3. Install the requirements

!pip install -r requirements.txt

Testing the model

Datasets and pretrained model during this part.

Sample run in the Drinks_detection.ipynb

!python3 test.py

Output metrics:

Training the model

Sample run in the Drinks_detection.ipynb .
If wandb visualization prompts , enter 3 (No visualization)

!python3 train.py

Output metrics after 50 epochs

Sample Video

Sample videos with object detection tracking is in the outputs folder Filename : output1.mp4

!python3 infer_record.py 

References:

1. TorchVision Object Detection Finetuning Tutoria
2. Pytorch Vision Reference
3. Object Detection Using Colab webcam