CueCAn: Cue Driven Contextual Attention Unit for Identifying Missing Traffic Signs on Uncontrained Roads

Dataset Download

Dataset Name: Missing Traffic Signs Video Dataset (MTSVD).

Available for download at the following link: https://idd.insaan.iiit.ac.in/dataset/download/

Conference Proceeding

International Conference on Robotics and Automation 2023 https://ieeexplore.ieee.org/document/10161576

Inference in Colab

https://colab.research.google.com/drive/1ujnqXKmQHUQP91COmuu_HXkjitzh0fGC?usp=sharing

Conda Environment Containing All Required Dependecies

conda env create --name fcnkeras --file=environment.yml

Missing Traffic Signs Video Dataset Samples

Missing Signs

Missing Traffic Signs Video Dataset (MTSVD)

Marked Missing Intervals

Traffic Sign Annotations and Tracks

CueCAn GradCam

Video Result

Running GradCam for Segmentation Task

Due to memory constraints in google colab, the gradCam for segmentation task can be run by the following:

git clone https://github.com/iHubData-Mobility/public-CueCAn.git
cd public-CueCAn
conda env create -f environment.yml
conda activate fcnkeras
gdown https://drive.google.com/uc?id=1xr0v0f-tMVCE-s_OCz5YCnkTXJW8HsS3
python gradSeg.py --model_dir cuecan_segmenter.h5 --img_dir images/context

And the segmentation outputs, would be stored in directory outSegGrad/