update

Author: dawnlh

Diff for: MetaSCI/.gitignore

@@ -1,7 +1,6 @@
 # zzh
 *.ckpt.*
 checkpoint
-dataset/*
 result/*
 ?bak/
 _bak/

Diff for: MetaSCI/README.md

@@ -9,13 +9,25 @@ This code is for CVPR 2021 paper "MetaSCI: Scalable and Adaptive Reconstruction
 ### Directory
 
 - `dataset\mask` : encoding mask, shape = [M N Cr NUM], i.e., NUM  [M N Cr] masks stacked along the last dimension.
-
-
+- `dataset\orig` : original frames, shape = [M N Cr], i.e., Cr  [M N] frames stacked along the last dimension.
 
 
 
 ### Parameter
 
 - `datadir`:  path for training set directory
-- `maskpath`: path for encoding mask  file
+- `maskpath`: path for encoding mask file
+
+
+
+### Run
+
+#### Training
+
+1. Prepare the masks: divide the large-scale mask into small patches (w./w.o. overlap) with the same size. Save one of the patch for the base model's training, and save the other patches for the adapting models' training.
+2. Train the base model with `main_MetaBaseModel_train.py` and one of the mask patch above (`main_MetaBaseModel_train_parallel.py` can be used for parallel training).
+3. Finetune the base model above to get the adapting models with `main_MetaAdaptModel_train.py` and the other mask patches.
+
+#### Test
 
+1. Use `main_MetaBaseModel_test.py` and `main_MetaAdaptModel_test.py ` and corresponding checkpoints and masks to reconstruct the video.