Update README.md
Browse files
README.md
CHANGED
|
@@ -3,8 +3,17 @@ license: mit
|
|
| 3 |
---
|
| 4 |
|
| 5 |
# Revisiting M3D is all you need
|
|
|
|
| 6 |
|
| 7 |
-
|
|
|
|
| 8 |
|
| 9 |
-
#
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
| 4 |
|
| 5 |
# Revisiting M3D is all you need
|
| 6 |
+
[**Reference Paper**](https://arxiv.org/abs/2404.00578) | [**Code**](https://github.com/babbu3682/AMOS-MM-MI2RL)
|
| 7 |
|
| 8 |
+
# Solution Overview
|
| 9 |
+
This code is the 3rd place solution for the MICCAI 2024 AMOS-MM Challenge. It is based on the M3D model and utilizes the LLaMA-3.1-Instruct-7B.
|
| 10 |
|
| 11 |
+
# Novelty
|
| 12 |
+
Our main innovations are as follows:
|
| 13 |
+
|
| 14 |
+
* We divided the body into three regions: chest, abdomen, and pelvis. This allowed us to focus more effectively on these areas and further improve performance through text prompt engineering.
|
| 15 |
+
* The introduction of LLaMA-3.1-8B-Instruct significantly boosted MRG and VQA performance.
|
| 16 |
+
* Additionally, the use of an error note strategy led to further improvements in the final results.
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
# Acknowledgement: We appreciate open source projects including: M3D, LLaVA.
|