Update README.md

README.md

@@ -1,6 +1,6 @@
 <div align="center">
 
-# SegMoTE: Token-Level Mixture of Experts for Medical Image Segmentation
+<h1 align="center">SegMoTE: Token-Level Mixture of Experts for Medical Image Segmentation</h1>
 
 **Yujie Lu<sup>1*</sup>, Jingwen Li<sup>2*</sup>, Sibo Ju<sup>3</sup>, Yanzhou Su<sup>4</sup>, He Yao<sup>1</sup>, Yisong Liu<sup>1</sup>, Min Zhu<sup>1&dagger;</sup>, Junlong Cheng<sup>1&dagger;</sup>
 
@@ -32,119 +32,6 @@ SegMoTE aims to retain the flexible prompt interface and generalization ability
 of SAM while providing lightweight adaptation for multimodal medical image
 segmentation.
 
-<p align="center">
-  <img src="figures/Introduction.png" width="98%" alt="SegMoTE motivation and comparison with previous SAM adaptation methods">
-</p>
-
-## Architecture
-
-<p align="center">
-  <img src="figures/SegMoTE.png" width="100%" alt="Overall architecture of SegMoTE">
-</p>
-
-SegMoTE extends SAM with two components. First, **MoTE** injects learnable
-expert tokens into the mask decoder and uses token-level routing to select
-specialized experts for different modalities and tasks. A load-balancing
-objective encourages effective expert utilization. Second, **PPT** pools image
-features into adaptive prompt tokens for selected few-class segmentation
-settings, reducing dependence on manual prompts. The framework supports point,
-bounding-box, and text prompts while retaining an efficient inference path.
-
-## Highlights
-
-- **Token-level expert routing:** SegMoTE activates expert tokens
-  for modality- and task-adaptive segmentation.
-- **Progressive prompt tokenization:** Feature-conditioned prompt tokens
-  support automatic segmentation for suitable binary foreground-background
-  tasks.
-- **Multimodal medical segmentation:** The framework is designed for medical
-  datasets spanning CT, MRI, dermoscopy, X-ray, and other modalities.
-- **SAM-compatible interaction:** Point, bounding-box, and text prompts are
-  supported in the released implementation.
-
-## Updates
-
-- **May 2026:** Code release preparation and inference checkpoint packaging.
-- **May 2026:** Deterministic MoTE routing enabled during evaluation.
-
-## Installation
-
-Create an environment with a CUDA-enabled PyTorch installation appropriate for
-your hardware, then install the remaining dependencies:
-
-```bash
-git clone <repository-url>
-cd SegMoTE
-pip install -r requirements.txt
-```
-
-The implementation uses PyTorch, TIMM, Transformers, MONAI, OpenCV, and common
-scientific Python packages.
-
-
-## Checkpoint
-
-Two checkpoint files are required for evaluation:
-
-```text
-checkpoints/sam_b.pth
-checkpoints/segmote.pth
-```
-
-| Checkpoint | Usage |
-| --- | --- |
-| `sam_b.pth` | Base initialization checkpoint loaded with `--sam_checkpoint` before loading SegMoTE weights. |
-| `segmote.pth` | SegMoTE inference checkpoint loaded with `--pretrain_path`. |
-
-Download the checkpoints from Baidu Netdisk:
-
-```text
-sam_b: https://pan.baidu.com/s/1HcmqPiwpWgnYr4CMf6Y9Pg  Password：eank
-segmote: https://pan.baidu.com/s/1tzlOv3YSU-9s6Gaw4pCF6g  Password：wja3
-```
-
-After downloading, place both checkpoint files in the `checkpoints/`
-directory. 
-
-## Evaluation
-
-Evaluate the released checkpoint on a dataset with bounding-box prompts:
-
-```bash
-python test.py \
-  --data_dir dataset \
-  --dataset_list BTCV \
-  --sam_checkpoint checkpoints/sam_b.pth \
-  --pretrain_path checkpoints/segmote.pth \
-  --prompt_mode bboxes \
-  --output_dir outputs/BTCV
-```
-
-## Training
-
-Train SegMoTE from the base initialization checkpoint:
-
-```bash
-python train.py \
-  --data_dir dataset \
-  --dataset_list BTCV \
-  --sam_checkpoint checkpoints/sam_b.pth \
-  --task_name segmote_train
-```
-
-For distributed training on multiple GPUs:
-
-```bash
-python train.py \
-  --data_dir dataset \
-  --dataset_list BTCV \
-  --sam_checkpoint checkpoints/sam_b.pth \
-  --task_name segmote_train \
-  --dist \
-  --multi_gpu \
-  --gpu_ids 0 1 2 3 4 5 6 7
-```
-
 
 ## Citation