Learning Semantic Boundaries from Noisy Annotations

STEAL

This is the official inference code for:

Devil Is in the Edges: Learning Semantic Boundaries from Noisy Annotations
David Acuna, Amlan Kar, Sanja Fidler

STEAL DEMO

Usage

Clone this repo

git clone https://github.com/nv-tlabs/STEAL
cd STEAL

Install dependencies

This code requires PyTorch 0.4 and python 3+. Please install dependencies by

pip install -r requirments.txt

Download pretrained models

Download the tar of the pretrained models from the Google Drive Folder, save it in 'checkpoints/', and run

cd checkpoints
tar -xvf checkpoints.tar.gz
cd ../

Inference (SBD)

python inference_sbd.py \
    --root_dir_val= ./data/sbd/data_aug/\
    --flist_val= ./data/sbd/data_aug/val_list.txt\
    --output_folder=./output/sbd/ \
    --ckpt=./checkpoints/sbd/model_checkpoint.pt\

Instructions and preprocessing scripts to download SBD and preprocess the dataset can be found here:
https://github.com/Chrisding/sbd-preprocess

Inference (Cityscapes)

python inference_cityscapes.py \
    --root_dir_val=./data/cityscapes-preprocess/data_proc \
    --flist_val=./data_proc/val.txt \
    --output_folder=./output/cityscapes/ \
    --ckpt=./checkpoints/cityscapes/model_checkpoint.pt\

Instructions and preprocessing scripts for Cityscapes can be found here:
https://github.com/Chrisding/cityscapes-preprocess

Test-NMS:
An example of how to apply TEST-NMS using Piotr's Structured Forest matlab toolbox. can be found in utils/edges_nms.m.
During training, we optimized for the same set of operations with r=2 (Check paper for more details)

Coarse-to-fine Demo

Checkout the ipython notebook that provides a simple walkthrough demonstrating how to run our model to refine coarsely annotated data.

Coarse to Fine

If you use this code, please cite:

@inproceedings{AcunaCVPR19STEAL,
title={Devil is in the Edges: Learning Semantic Boundaries from Noisy Annotations},
author={David Acuna and Amlan Kar and Sanja Fidler},
booktitle={CVPR},
year={2019}
}