This branch provides baseline code for few-shot object counting task.
VGG16Trans model is adopted for counting.
FSC147 dataset is released by LearningToCountEverything.
For more related works about class-agnostic counting(CAC), please refer this repository Awesome-Class-Agnostic-Counting
- fork or clone this repository
- pip install -r requirements.txt
- download dataset, then put images_384_VarV2 and gt_density_map_adaptive_384_VarV2 in ./datasets/FSC
- sh run.sh
| MAE/MSE | FSC-test | FSC-val |
|---|---|---|
| FamNet | 22.08/99.54 | 23.75/69.07 |
| VGG16Trans without examplar (baseline) | 17.51/132.62 | 19.21/77.43 |