turbo-boost-detection

turbo-boost-detection copied to clipboard

Turbo-boosting Object Detector

A PyTorch implementation, originally forked from a public repository based on the Mask-RCNN work.

Overview

• [ ] Better documentation
• [ ] Switch to PyTorch 0.4.x

For installation, please check the INSTALL documentation.

Test and Demo

TODO.

Train

See the script folder to get a sense of how to execuate train/evaluation commands in terminal.

The training schedule, learning rate, and other parameters can be set in the class object of CocoConfig in lib/config.py.

    sh script/base_8gpu.sh 105/meta_105_quick_1_roipool

Results

Results for bounding box and segmentation on COCO are reported based on the default configuration and backbone initialized with pretrained ImageNet weights. The metric is mAP on IoU=0.50:0.95.

TODO: more results to come

Misc

To transfer from server to local machine:

    scp -r s42:/DATA/hyli/project/turbo-boost-detection/results/your_folder .