Depth Map Generation
Name: “depth_map_generation”
Description
Generates a depth map from a single image using the MoGe2 model.
Warning
This application is only available if the CARS Edge Detection plugin <https://github.com/CNES/cars-edge-detection-plugin> is installed.
Configuration
Name |
Description |
Type |
Available value |
Default value |
Required |
|---|---|---|---|---|---|
method |
Method for depth map generation |
string |
“moge2” |
“moge2” |
No |
model |
Name of the model to use for depth map generation |
string |
“Ruicheng/moge-2-vitl-normal” |
No |
|
edge_threshold |
Threshold for edge detection |
float |
should be >= 0, <= 1 |
0.6 |
No |
save_intermediate_data |
Save epipolar images and texture |
boolean |
false |
No |
Model
The model parameter may be any MoGe2 model available on the HuggingFace hub, or any path to a local model.
Officially supported models are:
Ruicheng/moge-2-vitl-normal
Ruicheng/moge-2-vitb-normal
Ruicheng/moge-2-vits-normal
If a model is not already available locally, it will be downloaded from the HuggingFace hub and cached for future use.
The executable cars-download-moge2 can be used to download a model, as further explained in the Edge detection example section.
Example
--- applications: depth_map_generation: model: "Ruicheng/moge-2-vitl-normal" edge_threshold: 0.4 # lower values will result in more edges being detected save_intermediate_data: false
{ "applications": { "depth_map_generation": { "model": "Ruicheng/moge-2-vitl-normal", "edge_threshold": 0.4, "save_intermediate_data": false } } }