CURE-TSD: Challenging Unreal and Real Environment for Traffic Sign Detection

As one of the research directions at OLIVES Lab @ Georgia Tech, we focus on the robustness of data-driven algorithms under diverse challenging conditions where trained models can possibly be depolyed. To achieve this goal, we introduced a large-sacle (~1.72M frames) traffic sign detection video dataset (CURE-TSD) which is among the most comprehensive datasets with controlled synthetic challenging conditions. The video sequences in the CURE-TSD dataset are grouped into two classes: real data and unreal data. Real data correspond to processed versions of sequences acquired from real world. Unreal data corresponds to synthesized sequences generated in a virtual environment. There are 49 real sequences and 49 unreal sequences that do not include any specific challenge. We separated the sequences into 70% and 0 splits. Therefore, we have 34 training videos and 15 test videos in both real and unreal sequences that are challenge-free. There are 300 frames in each video sequence. There are 49 challenge-free real video sequences processed with 12 different types of effects and 5 different challenge levels, which result in 2,989 (49125+49) video sequences. Moreover, there are 49 synthesized video sequences processed with 11 different types of effects and 5 different challenge levels, which leads to 2,744 (49115+49) video sequences. In total, there are 5,733 video sequences, which include around 1.72 million frames. Please refer to our GitHub page for code, papers, and more information.