The target of this homework is object detection.
Training on SVHN dataset with digits in different types and trying to detect and recognize number zero to nine.
I use Faster-RCNN as a pretrained model.
The following specs were used to create the original solution.
- Ubuntu 18.04 LTS
- NVIDIA GeForce RTX 2080
You can download the training and testing dataset from Google Drive.
https://drive.google.com/drive/folders/1VHW53unZJoPN8Bu1aN4FUq6SXAT60lOJ?usp=sharing
- Linux or macOS with Python ≥ 3.6
- PyTorch ≥ 1.5 and torchvision that matches the PyTorch installation.
- OpenCV is optional and needed by demo and visualization
- Numpy 1.19.2
- Tqdm 4.51.0
- Cuda 10.1
- Detectron 2
You need to create a dictory names 'checkpoints' to save chekpoint.
Download Detectron2 by follow the github
https://github.com/facebookresearch/detectron2
Run train.py to start training.
python train.py
Test the model
python Test.py
https://papers.nips.cc/paper/2015/file/14bfa6bb14875e45bba028a21ed38046-Paper.pdf
http://ufldl.stanford.edu/housenumbers/
https://github.com/facebookresearch/detectron2