- It is better to use https://github.com/lgvaz/mantisshrimp
- All the features in this (plus extra features) have been added to mantisshrimp.
- It is a complete framework for object detection and more scalable.
Faster RCNN Fine-Tune Implementation in Pytorch.
- git clone the repo
git clone https://github.com/oke-aditya/pytorch_fasterrcnn.git
- install the requirements (will add)
pip install -r requirements.txt
Simply edit the config file to set your hyper parameters.
- Keep the training and validation csv file as follows
NOTE
Do not use target as 0 class. It is reserved as background.
image_id xtl ytl xbr ybr target
1 xmin ymin xmax ymax 1
1 xmin ymin xmax ymax 2
2 xmin ymin xmax ymax 3
-
Simply edit the config file to set your hyper parameters
-
Run the train.py file
- It works for multiple class object detection.
-
Note that backbones are pretrained on imagenet.
-
Following backbones are supported
- vgg11, vgg13, vgg16, vgg19
- resnet18, resnet34, resnet50, resnet101, resnet152
- renext101
- mobilenet_v2
Sample Outputs
If you like the implemenation or have taken an inspiration do give a star :-)