Kaggle IMaterialist

Solution

This repo is based on Matterport MaskRCNN + DenseCRF for post processing

Using Dense CRF makes the segmentation look better
In order to use densecrf, you need to install the repo in the link https://github.com/lucasb-eyer/pydensecrf
Densecrf requires the probability of each pexel belonging to each class, whereas this library only returns 1 or 0
so change some files with the files in the "modified" folder

Name		Name	Last commit message	Last commit date
Latest commit History 213 Commits
assets		assets
images		images
modified		modified
mrcnn		mrcnn
samples		samples
.gitignore		.gitignore
Baseline MaskRCNN.ipynb		Baseline MaskRCNN.ipynb
Baseline MaskRCNN.py		Baseline MaskRCNN.py
Inference_fashion.ipynb		Inference_fashion.ipynb
LICENSE		LICENSE
MANIFEST.in		MANIFEST.in
README.md		README.md
Untitled.ipynb		Untitled.ipynb
densecrf.png		densecrf.png
maskrcnn.png		maskrcnn.png
requirements.txt		requirements.txt
sample_submission.csv		sample_submission.csv
setup.cfg		setup.cfg
setup.py		setup.py