In the last few years, many algorithms with remarkable effectiveness for Object Detection have been published but still some comparative metrics haven’t been defined.
The difficulties in making this comparison arise from the fact that different algorithms are based on different Feature Extractors (VGGs, Residual Networks, etc.), different base resolution and different implementation on specific platforms.
For this study Addfor S.p.A. will provide advanced technical support and computing platforms available at the moment:
Training Platforms: TITAN Xp, GTX1080 Ti and IBM PowerAI NVIDIA P100 / V100 servers.
Inference Platforms: NVIDIA Drive PX2 and Intel Movidius.
The goal of this thesis is to define objective methods for measuring the performance of image and video understanding algorithms in real datasets. The first will be a research work on state-of-the-art for Object Detection and Image Segmentation.
The second part of this dissertation project will be devoted to the study of existing metrics for the evaluation and comparison of these algorithms in terms of accuracy and time of calculation.
Addfor will provide data sets of real-time image sequences obtained in different weather conditions and in non-optimal conditions (flickering and jittering) that represent the common difficulties found on real computer vision tasks.
The final purpose of the thesis will be to define a workflow for testing these new algorithms and evaluating them on similar real datasets.
Who we’re looking for
Students that are about to get their master degree in: Computer Science, Mathematics
Skills: Python, C/C++, math, exp. w. at least one among TensorFlow / PyTorch / BigDL
Duration of this Projects: 6-8 months