update finetune to README.md

Author: Mark-ZhouWX

research/segment-anything/README.md

@@ -68,13 +68,62 @@ Since SAM can efficiently process prompts, masks for the entire image can be gen
 python use_sam_with_amg.py --model-type vit_h
 ```
 
-<p float="left">
-<img alt="img.png" width="400" src="images/dengta.jpg"/>
-<img alt="img.png" width="400" src="images/dengta-amg-vith.png"/>
-</p>
+<div align="center">
+<img src="images/dengta.jpg" height="350" />
+      
+<img src="images/dengta-amg-vith.png" height="350" />
+</div>
 
 See `python use_sam_with_amg.py --help` to explore more custom settings.
 
 ## Finetune
 
-To be continued
+Finetune is a popular method that adapts large pretrained model to specific downstream tasks. Currently, finetune with box-prompt are supported. The bounding boxes are used as prompt input to predict mask.
+Beside fine-tuning our code on COCO2017 dataset which contains common seen objects and lies in the similar distribution of the original [training dataset](https://segment-anything.com/dataset/index.html)) of SAM, We have done further experiments on a medical imaging segmentation dataset [FLARE22](https://flare22.grand-challenge.org/Dataset/). Result shows that the finetune method in this repository is effective.
+
+The bellowing shows the mask quality before and after finetune.
+
+
+| pretrained_model | dataset  |    epochs    | mIOU |
+| :--------------: | -------- | :-----------: | ---- |
+|    sam-vit-b    | COCO2017 | 0 (zero-shot) | 77.4 |
+|    sam-vit-b    | COCO2017 |      20      | 83.6 |
+|    sam-vit-b    | FLARE22  | 0 (zero-shot) | 79.5 |
+|    sam-vit-b    | FLARE22  |      10      | 88.1 |
+
+To finetune COCO dataset, please run:
+
+```shell
+mpirun --allow-run-as-root -n 8 python train.py -c configs/coco_box_finetune.yaml
+```
+
+The original FLARE22 dataset contains image in 3D format and ground truth labelled as instance segmentation ids. Run
+
+```shell
+python scripts/preprocess_CT_MR_dataset.py
+```
+
+to preprocess it to the format of 2D RGB image and binary mask
+
+To finetune FLARE22 dataset, please run:
+
+```shell
+mpirun --allow-run-as-root -n 8 python train.py -c configs/flare_box_finetune.yaml
+```
+
+Here are the examples of segmentation result predicted by fine-tuned SAM:
+
+<div align="center">
+<img src="images/coco_bear.jpg" height="350" />
+    
+<img src="images/flare_organ.jpg" height="350" />
+</div>
+
+<p align="center">
+  <em> COCO2017 image example</em>
+                        
+                        
+  <em> FLARE22 image example </em>
+</p>
+
+