Sinovation Ventures Challenge
Please refer to install.md for installation and dataset preparation.
Please see getting_started.md for the basic usage of MMDetection.
Link the dataset to the folder "./data/" under this repo.
ln -s path/to/dataset data
./tools/dist_train.sh local_config/atss_r50_fpn_ms12.py 8
Note that the model of this example is trained under 8-GPU settings. If you donot have enough GPU cards, try to adjust some related settings for efficient training.
Run test_example.ipynb to see detected results (see as follows) of pretrained baseline model. The baseline model is relatively weak, try your best to improve it~!
./tools/dist_test.sh local_config/atss_r50_fpn_ms12.py pretrain_model/atss_r50_fpn_ms12.model 8 --format-only --options "jsonfile_prefix=./submit_atss_r50_fpn_ms12_results"
Pay attention to the constraints of the complexity for your detectors. The following commands are used for official judgements, with 640x400 images for inference.
python3 ./tools/get_flops.py local_config/atss_r50_fpn_ms12.py --shape 640 400
python3 ./tools/benchmark.py local_config/atss_r50_fpn_ms12.py pretrain_model/atss_r50_fpn_ms12.model --fuse-conv-bn
MMDetection is an open source project that is contributed by researchers and engineers from various colleges and companies. The technical report is on ArXiv.
Documentation: https://mmdetection.readthedocs.io/
The branch works with PyTorch 1.3 to 1.5.
This project is released under the Apache 2.0 license.