PSP-HDRI$+$: A Synthetic Dataset Generator for Pre-Training of Human-Centric Computer Vision Models
Ebadi, Salehe Erfanian, Dhakad, Saurav, Vishwakarma, Sanjay, Wang, Chunpu, Jhang, You-Cyuan, Chociej, Maciek, Crespi, Adam, Thaman, Alex, Ganguly, Sujoy
–arXiv.org Artificial Intelligence
Consequently, exploiting synthetic data HDRI+ that proves to be a superior pre-training to its full potential is yet to be realized. Motivated by alternative to ImageNet and other large-scale synthetic these limitations, we present a privacy-preserving, ethically data counterparts. We demonstrate that sourced, and fully manipulable synthetic data generator for pre-training with our synthetic data will yield a human-centric computer vision named PSP-HDRI+, which more general model that performs better than alternatives is built upon PeopleSansPeople (Ebadi et al., 2021) (PSP) even when tested on out-of-distribution in Unity
arXiv.org Artificial Intelligence
Jul-11-2022
- Genre:
- Research Report (0.40)
- Technology:
- Information Technology > Artificial Intelligence
- Machine Learning (1.00)
- Vision (1.00)
- Information Technology > Artificial Intelligence