Team Member: Cai Fengyu, Zhou Wanhao, Sepehri Yamin
Report of the result on the test @IOU=0.3
- detection by segmentation: 0.29
- detection by sliding window and classifier: 0.23
- detection by Mask-RCNN: 0.71
Report of the result on the competition
- Group Name: Group2
- Best F1 score: 0.734 (ranked 2nd place in the competation)
Sorry for the inconvenience that we have submissions from two team names: 'Group2' and 'Group2 Method3'. Please, just ignore the second one! Thank you so much!
Packages:
- keras
- tensorflow
- skimage
- scipy
- sklearn
- numpy Hardware:
- GPU Tesla K80 with 15G memory
- 12G memory provided by Colab
- Get the module of Mask-RCNN from the github
$ git clone --quiet https://github.com/matterport/Mask_RCNN.git $ cd Mask_RCNN $ pip install -q PyDrive $ pip install -r requirements.txt $ python setup.py install - Get the pretrained weight based on the coco dataset
- If you would like to skip the training process, you can also download our trained weights from here
- Mask R-CNN for Object Detection and Segmentation， https://github.com/matterport/Mask_RCNN
- Lecture Notes, EE451 Image Analysis and Pattern Recognition, Spring 2019
- Watershedding algorithm, https://en.wikipedia.org/wiki/Watershed_(image_processing)