RoCoSDF / Git / Diff of /README.md

Models:

SCallahan/

RoCoSDF

Downloads: 1

Diff of /README.md [98712b] .. [8e7194]

Switch to unified view


## RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D ultrasound Imaging Shape Reconstruction
--------------------------------------

The official implementation code for MICCAI 2024 paper:
[RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction](https://chenhbo.github.io/RoCoSDF/)
by [Hongbo Chen](https://chenhbo.github.io/), Yuchong Gao, Shuhang Zhang, Jiangjie Wu, [Yuexin Ma](https://yuexinma.me/) and [Rui Zheng](https://sist.shanghaitech.edu.cn/zhengrui_en/main.htm).


RoCoSDF is a framework built on neural implicit signed distance functions for shape reconstruction of multi-view freehand 3D ultrasound imaging.

<div align="center">
<img src="https://github.com/chenhbo/RoCoSDF/blob/main/img/Fig_RoCoScan.png?raw=true" style="zoom:15%" alt="Data Aquisition Protocol"/>
</div>




## Demo
* Thoracic Vertebra T4 from ultrasound transducer 1 (UT1)

<div align="center">
<img src="https://github.com/chenhbo/RoCoSDF/blob/main/img/Fig_Result_T4.png?raw=true" style="zoom:14.7%" alt="Framework"/>
</div>


<br />


* The example mesh results of RoCoSDF are in `outs/T4_RoCo/outputs/*.ply`.


--------------------------------------

## Usage
Our code is implemented in NVIDIA 3090, Ubuntu 18/20, Python 3.8, PyTorch 1.12.1 and CUDA 11.6.


### Install Dependencies 
For 20/30x GPU:
```
conda create -n rocosdf python=3.8
conda activate rocosdf
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
pip install tqdm pyhocon==0.3.57 trimesh PyMCubes scipy matplotlib
pip install visdom open3d scikit-image plyfile
```

For 40x GPU, cuda 11.8:
```
conda create -n rocosdf python=3.10
conda activate rocosdf
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install tqdm pyhocon==0.3.57 trimesh PyMCubes scipy matplotlib
pip install visdom open3d scikit-image plyfile
```
--------------------------------------

### Data Preparation
- Convert the row-scan and column-scan segmented volumetric mask to point clouds file *.ply.

- Both the row-scan and column-scan point clouds should be in a same tracking 
space or manually aligned in a unified space.

- Put the row-scan and column-scan point clouds data in ./data.

```
RoCoSDF/
│
├── data/
│   ├── T4_Co.ply            % your own data
|   ├── T4_Ro.ply            % your own data
│   ├── T4_Co_ds.pt          % generated during data preprocessing, downsampled point clouds for training
|   ├── T4_Ro_ds.pt          % generated during data preprocessing, downsampled point clouds for training
│   ├── T4_Co_sampler.pt     % generated during training
|   ├── T4_Ro_sampler.pt     % generated during training
|
|
├── outs/
│   ├── T4_Co/
│       └── outputs/
|           └── *.ply
│   ├── T4_Ro/
│       └── outputs/
|            └── *.ply
│   ├── T4_RoCo/
│       └── outputs/
|            └── *.ply
|
```

--------------------------------------

### Run RoCoSDF
In Linux, directly train the model through `sh train.sh` OR using command as below.

```
python runRoCoSDF.py --gpu 0  --conf confs/conf.conf --dataname T4_Co --dataname2 T4_Ro  --dir T4_Co --dir2 T4_Ro --dir3 T4_RoCo --mode train
 ```


### Run SDF Refinement Only
In Linux, directly train the model through `sh train_refine_only.sh` OR using command as below.

```
python runRoCoSDF.py --gpu 0  --conf confs/conf.conf --dataname T4_Co --dataname2 T4_Ro --dir T4_Co --dir2 T4_Ro --dir3 T4_RoCo --mode train_refine
 ```

### Contact
For any queries, please contact [chenhb[at]shanghaitech.edu.cn](mailto:chenhb@shanghaitech.edu.cn).

### Citation
If you use RoCoSDF in your research, please cite the paper:

```
 @InProceedings{chenRoCoSDF,
   author="Chen, Hongbo
   and Gao, Yuchong
   and Zhang, Shuhang
   and Wu, Jiangjie
   and Ma, Yuexin
   and Zheng, Rui",
   title="RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction",
   booktitle="Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024",
   year="2024",
   publisher="Springer Nature Switzerland",
   address="Cham",
   pages="721--731",
   isbn="978-3-031-72083-3"
   }
```
--------------------------------------

### References
The reference codes are from the following links.
We appreciate all the contributors.

* FUNSR(UNSR in the proceeding): https://github.com/chenhbo/FUNSR
  
* DeepSDF: https://github.com/facebookresearch/DeepSDF

* NeuralPull: https://github.com/mabaorui/NeuralPull-Pytorch

* GenSDF: https://github.com/princeton-computational-imaging/gensdf

* CSGSDF: https://github.com/zoemarschner/csg_on_nsdf

	a/README.md		b/README.md
1	## RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D ultrasound Imaging Shape Reconstruction	1	## RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D ultrasound Imaging Shape Reconstruction
2	--------------------------------------	2	--------------------------------------
3		3
4	The official implementation code for MICCAI 2024 paper:	4	The official implementation code for MICCAI 2024 paper:
5	[RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction](https://chenhbo.github.io/RoCoSDF/)	5	[RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction](https://chenhbo.github.io/RoCoSDF/)
6	by [Hongbo Chen](https://chenhbo.github.io/), Yuchong Gao, Shuhang Zhang, Jiangjie Wu, [Yuexin Ma](https://yuexinma.me/) and [Rui Zheng](https://sist.shanghaitech.edu.cn/zhengrui_en/main.htm).	6	by [Hongbo Chen](https://chenhbo.github.io/), Yuchong Gao, Shuhang Zhang, Jiangjie Wu, [Yuexin Ma](https://yuexinma.me/) and [Rui Zheng](https://sist.shanghaitech.edu.cn/zhengrui_en/main.htm).
7		7
8		8
9	RoCoSDF is a framework built on neural implicit signed distance functions for shape reconstruction of multi-view freehand 3D ultrasound imaging.	9	RoCoSDF is a framework built on neural implicit signed distance functions for shape reconstruction of multi-view freehand 3D ultrasound imaging.
10		10
11	<div align="center">	11	<div align="center">
12	<img src="img/Fig_RoCoScan.png" style="zoom:15%" alt="Data Aquisition Protocol"/>	12	<img src="https://github.com/chenhbo/RoCoSDF/blob/main/img/Fig_RoCoScan.png?raw=true" style="zoom:15%" alt="Data Aquisition Protocol"/>
13	</div>	13	</div>
14		14
15		15
16		16
17		17
18	## Demo	18	## Demo
19	* Thoracic Vertebra T4 from ultrasound transducer 1 (UT1)	19	* Thoracic Vertebra T4 from ultrasound transducer 1 (UT1)
20		20
21	<div align="center">	21	<div align="center">
22	<img src="img/Fig_Result_T4.png" style="zoom:14.7%" alt="Framework"/>	22	<img src="https://github.com/chenhbo/RoCoSDF/blob/main/img/Fig_Result_T4.png?raw=true" style="zoom:14.7%" alt="Framework"/>
23	</div>	23	</div>
24		24
25		25
26	<br />	26	<br />
27		27
28		28
29	* The example mesh results of RoCoSDF are in `outs/T4_RoCo/outputs/*.ply`.	29	* The example mesh results of RoCoSDF are in `outs/T4_RoCo/outputs/*.ply`.
30		30
31		31
32	--------------------------------------	32	--------------------------------------
33		33
34	## Usage	34	## Usage
35	Our code is implemented in NVIDIA 3090, Ubuntu 18/20, Python 3.8, PyTorch 1.12.1 and CUDA 11.6.	35	Our code is implemented in NVIDIA 3090, Ubuntu 18/20, Python 3.8, PyTorch 1.12.1 and CUDA 11.6.
36		36
37		37
38	### Install Dependencies	38	### Install Dependencies
39	For 20/30x GPU:	39	For 20/30x GPU:
40	```	40	```
41	conda create -n rocosdf python=3.8	41	conda create -n rocosdf python=3.8
42	conda activate rocosdf	42	conda activate rocosdf
43	conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge	43	conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
44	pip install tqdm pyhocon==0.3.57 trimesh PyMCubes scipy matplotlib	44	pip install tqdm pyhocon==0.3.57 trimesh PyMCubes scipy matplotlib
45	pip install visdom open3d scikit-image plyfile	45	pip install visdom open3d scikit-image plyfile
46	```	46	```
47		47
48	For 40x GPU, cuda 11.8:	48	For 40x GPU, cuda 11.8:
49	```	49	```
50	conda create -n rocosdf python=3.10	50	conda create -n rocosdf python=3.10
51	conda activate rocosdf	51	conda activate rocosdf
52	pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118	52	pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
53	pip install tqdm pyhocon==0.3.57 trimesh PyMCubes scipy matplotlib	53	pip install tqdm pyhocon==0.3.57 trimesh PyMCubes scipy matplotlib
54	pip install visdom open3d scikit-image plyfile	54	pip install visdom open3d scikit-image plyfile
55	```	55	```
56	--------------------------------------	56	--------------------------------------
57		57
58	### Data Preparation	58	### Data Preparation
59	- Convert the row-scan and column-scan segmented volumetric mask to point clouds file *.ply.	59	- Convert the row-scan and column-scan segmented volumetric mask to point clouds file *.ply.
60		60
61	- Both the row-scan and column-scan point clouds should be in a same tracking	61	- Both the row-scan and column-scan point clouds should be in a same tracking
62	space or manually aligned in a unified space.	62	space or manually aligned in a unified space.
63		63
64	- Put the row-scan and column-scan point clouds data in ./data.	64	- Put the row-scan and column-scan point clouds data in ./data.
65		65
66	```	66	```
67	RoCoSDF/	67	RoCoSDF/
68	│	68	│
69	├── data/	69	├── data/
70	│ ├── T4_Co.ply % your own data	70	│ ├── T4_Co.ply % your own data
71	\| ├── T4_Ro.ply % your own data	71	\| ├── T4_Ro.ply % your own data
72	│ ├── T4_Co_ds.pt % generated during data preprocessing, downsampled point clouds for training	72	│ ├── T4_Co_ds.pt % generated during data preprocessing, downsampled point clouds for training
73	\| ├── T4_Ro_ds.pt % generated during data preprocessing, downsampled point clouds for training	73	\| ├── T4_Ro_ds.pt % generated during data preprocessing, downsampled point clouds for training
74	│ ├── T4_Co_sampler.pt % generated during training	74	│ ├── T4_Co_sampler.pt % generated during training
75	\| ├── T4_Ro_sampler.pt % generated during training	75	\| ├── T4_Ro_sampler.pt % generated during training
76	\|	76	\|
77	\|	77	\|
78	├── outs/	78	├── outs/
79	│ ├── T4_Co/	79	│ ├── T4_Co/
80	│ └── outputs/	80	│ └── outputs/
81	\| └── *.ply	81	\| └── *.ply
82	│ ├── T4_Ro/	82	│ ├── T4_Ro/
83	│ └── outputs/	83	│ └── outputs/
84	\| └── *.ply	84	\| └── *.ply
85	│ ├── T4_RoCo/	85	│ ├── T4_RoCo/
86	│ └── outputs/	86	│ └── outputs/
87	\| └── *.ply	87	\| └── *.ply
88	\|	88	\|
89	```	89	```
90		90
91	--------------------------------------	91	--------------------------------------
92		92
93	### Run RoCoSDF	93	### Run RoCoSDF
94	In Linux, directly train the model through `sh train.sh` OR using command as below.	94	In Linux, directly train the model through `sh train.sh` OR using command as below.
95		95
96	```	96	```
97	python runRoCoSDF.py --gpu 0 --conf confs/conf.conf --dataname T4_Co --dataname2 T4_Ro --dir T4_Co --dir2 T4_Ro --dir3 T4_RoCo --mode train	97	python runRoCoSDF.py --gpu 0 --conf confs/conf.conf --dataname T4_Co --dataname2 T4_Ro --dir T4_Co --dir2 T4_Ro --dir3 T4_RoCo --mode train
98	```	98	```
99		99
100		100
101	### Run SDF Refinement Only	101	### Run SDF Refinement Only
102	In Linux, directly train the model through `sh train_refine_only.sh` OR using command as below.	102	In Linux, directly train the model through `sh train_refine_only.sh` OR using command as below.
103		103
104	```	104	```
105	python runRoCoSDF.py --gpu 0 --conf confs/conf.conf --dataname T4_Co --dataname2 T4_Ro --dir T4_Co --dir2 T4_Ro --dir3 T4_RoCo --mode train_refine	105	python runRoCoSDF.py --gpu 0 --conf confs/conf.conf --dataname T4_Co --dataname2 T4_Ro --dir T4_Co --dir2 T4_Ro --dir3 T4_RoCo --mode train_refine
106	```	106	```
107		107
108	### Contact	108	### Contact
109	For any queries, please contact [chenhb[at]shanghaitech.edu.cn](mailto:chenhb@shanghaitech.edu.cn).	109	For any queries, please contact [chenhb[at]shanghaitech.edu.cn](mailto:chenhb@shanghaitech.edu.cn).
110		110
111	### Citation	111	### Citation
112	If you use RoCoSDF in your research, please cite the paper:	112	If you use RoCoSDF in your research, please cite the paper:
113		113
114	```	114	```
115	@InProceedings{chenRoCoSDF,	115	@InProceedings{chenRoCoSDF,
116	author="Chen, Hongbo	116	author="Chen, Hongbo
117	and Gao, Yuchong	117	and Gao, Yuchong
118	and Zhang, Shuhang	118	and Zhang, Shuhang
119	and Wu, Jiangjie	119	and Wu, Jiangjie
120	and Ma, Yuexin	120	and Ma, Yuexin
121	and Zheng, Rui",	121	and Zheng, Rui",
122	title="RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction",	122	title="RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction",
123	booktitle="Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024",	123	booktitle="Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024",
124	year="2024",	124	year="2024",
125	publisher="Springer Nature Switzerland",	125	publisher="Springer Nature Switzerland",
126	address="Cham",	126	address="Cham",
127	pages="721--731",	127	pages="721--731",
128	isbn="978-3-031-72083-3"	128	isbn="978-3-031-72083-3"
129	}	129	}
130	```	130	```
131	--------------------------------------	131	--------------------------------------
132		132
133	### References	133	### References
134	The reference codes are from the following links.	134	The reference codes are from the following links.
135	We appreciate all the contributors.	135	We appreciate all the contributors.
136		136
137	* FUNSR(UNSR in the proceeding): https://github.com/chenhbo/FUNSR	137	* FUNSR(UNSR in the proceeding): https://github.com/chenhbo/FUNSR
138		138
139	* DeepSDF: https://github.com/facebookresearch/DeepSDF	139	* DeepSDF: https://github.com/facebookresearch/DeepSDF
140		140
141	* NeuralPull: https://github.com/mabaorui/NeuralPull-Pytorch	141	* NeuralPull: https://github.com/mabaorui/NeuralPull-Pytorch
142		142
143	* GenSDF: https://github.com/princeton-computational-imaging/gensdf	143	* GenSDF: https://github.com/princeton-computational-imaging/gensdf
144		144
145	* CSGSDF: https://github.com/zoemarschner/csg_on_nsdf	145	* CSGSDF: https://github.com/zoemarschner/csg_on_nsdf