file_parsing

[mountain111]

1 base_classes**

fsdet/data/datasets/builtin_meta.py

2 directory of novel_classes_fine_tune

fsdet/data/builtin.py fsdet/data/datasets/meta_pascal_voc.py (1)seed0 voc_2007_trainval_novel1_1shot (2)seedx, x>=1 voc_2007_trainval_novel1_1shot_seedx

3 split

fsdet/data/datasets/builtin_meta.py

4 split vs. seed

split: base classes and novel classes, 1, 2, 3, i.e., novel1, novel2, novel3 seed: different novel samples from novel classes

5 datasets

(1) register in fsdet/data/datasets/builtin.py (2) change for get_detection_dataset_dicts in fsdet/data/build.py ​ filter_image in fsdet/data/build.py

6 output layer change

(1) in fsdet/modeling/roi_heads/fast_rcnn.py change output layer

​ import math in fsdet/modeling/roi_heads/fast_rcnn.py

(2) in fsdet/modeling/roi_heads/roi_heads.py ​ MetadataCatalog, self.output_layer, dataset_names

7 similarity

(1) fsdet/data/datasets/pascal_voc.py

8 fix seed

fsdet/utils/env.py
fsdet/config/defaults.py

9 sampling

fsdet/data/samplers/distributed_sampler.py