DLR-RM / BlenderProc

A procedural Blender pipeline for photorealistic training image generation
GNU General Public License v3.0
2.64k stars 438 forks source link

During the simulation process, only local information of the object is recognized for annotation #1122

Open Bainily028 opened 5 days ago

Bainily028 commented 5 days ago

Describe the issue

Hello, Dear developers,, thank you again for your contribution. I have a question I would like to ask while using BlendPro. Can I only focus on the local information of the object and generate a mask for that local information during the process of creating a dataset for blenderproc simulation. For example, the following model is a combination of a cube and a cone. In the simulation process, I only want to focus on the cone part and generate a mask with only the cone part. May I ask if the current blenderproc supports this feature? image

Minimal code example

No response

Files required to run the code

No response

Expected behavior

Only focus on local information of the model during the simulation process.

BlenderProc version

V2.7.1

cornerfarmer commented 5 days ago

I dont understand your question. Could you please provide a minimal code example and describe what output you would like.

Bainily028 commented 5 days ago

Hey@cornerfarmer ,Thank you for your reply. Every question is always answered promptly.

Sorry, maybe my description was not clear enough.

Let me explain it in a different scenario. I want to use BlendPro to complete my segmentation task.

For example, the objects and scenes below are the results of my simulation using BlendPro. In the scene, I can obtain the mask of the entire object of the bottle, but if my core goal is not the entire bottle, but a partial part of the bottle, such as the bottle cap. The final output mask is only the mask of the bottle cap, not the mask of the entire bottle.

So, can I achieve this in BlendProc by importing the overall model for simulation, but only outputting the mask at the specified position in the model.

image