Code release for the paper "Probing Contextual Diversity for Dense Out-of-Distribution Detection"

This repository contains evaluation code and pre-trained models.

Setup

• Install required packages. The code has been tested with PyTorch 1.9 / Python 3.8: pip install -r requirements.txt

• Create folders for datasets and model snapshots: mkdir -p checkpoints datasets/street_hazards

Datasets

Download and extract the desired dataset:

• StreetHazards:
  • wget https://people.eecs.berkeley.edu/~hendrycks/streethazards_test.tar -P datasets/street_hazards/
  • tar -xf datasets/street_hazards/streethazards_test.tar -C datasets/street_hazards/
• RoadAnomaly:
  • wget https://datasets-cvlab.epfl.ch/2019-road-anomaly/RoadAnomaly_jpg.zip -P datasets/
  • unzip datasets/RoadAnomaly_jpg.zip -d datasets/

Model Checkpoints

Download the desired checkpoint and place it in the checkpoints folder:

• StreetHazards: https://drive.google.com/file/d/1mdTDlOZTjWf1YAIHYiuUFCGKgC4N7KTC/view?usp=sharing
• RoadAnomaly: https://drive.google.com/file/d/18NbxWfKnxpRsyB9CySGFB7-pXjLksy5y/view?usp=sharing

Dataset Evlauation

Note: since the results reported in the paper are averages over multiple runs, the outcomes obtained with this code will differ slightly.

• StreetHazards: python eval.py street_hazards checkpoints/StreetHazards_deeplabv3_resnet50.pth --arch deeplabv3_resnet50
• RoadAnomaly: python eval.py road_anomaly checkpoints/BDD100k_deeplabv3plus_resnet101.pth --arch deeplabv3plus_resnet101