1. mmcv is an open source package.
2. models contains different models we used.

   For noise robustness task, FineNet adopt "CORnet_TwopassA",
       Details of CoarseNet and FineNet structures are found in  scripts/configs/CORnet_RT_Twopass_config.py

   For FFL,  model is adopt "CORnet_TwopassB", distill_fused=1, distill_ensem=1
       Details of CoarseNet and FineNet structures are found in  scripts/configs/CORnet_RT_ensemble_config.py


   For SFL,  model is adopt "CORnet_TwopassB",distill_fused=0, distill_ensem=0
       Details of CoarseNet and FineNet structures are found in  scripts/configs/CORnet_RT_ensemble_config.py



   trianing two-pathway model:
   python train_twopass_cache_iter.py CORnet_RT_Twopass_config.py

   training FFL model :
   python train_twopass_ensemble.py CORnet_RT_ensemble_config.py --distill_fused 1 --distill_ensem 1

   training FFL model :
   python train_twopass_ensemble.py CORnet_RT_ensemble_config.py --distill_fused 0 --distill_ensem 0
