Closed leslie-ds closed 2 years ago
Just use datasets shown in WACV paper using default settings
The RGB stream has identical architecture as HRNet, which is a great network for learning fine details in the RGB domain. The reasons for choosing HRNet as the backbone is illustrated in both papers.
How to train a model consistent with CAT_full_v1 performance?