Thanks for your brilliant work first. In the official repo, there're no demos on how to editting the gaussians with text-prompts. (Only an object id for removal/inpainting tasks), However, this would sometimes be hard to utilize cause one cannot know what exactly the object is given an id of that object. Thus I did a small change in the "prepare pseudo label" stage. By using gaussian-grouping/Tracking-Anything-with-DEVA/demo/demo_with_text.py, we can input text prompts to get object masks as you want. With this mask accuired by prompts, we can then do the training, removal/inpainting , any downstream tasks.
Thanks for your brilliant work first. In the official repo, there're no demos on how to editting the gaussians with text-prompts. (Only an object id for removal/inpainting tasks), However, this would sometimes be hard to utilize cause one cannot know what exactly the object is given an id of that object. Thus I did a small change in the "prepare pseudo label" stage. By using gaussian-grouping/Tracking-Anything-with-DEVA/demo/demo_with_text.py, we can input text prompts to get object masks as you want. With this mask accuired by prompts, we can then do the training, removal/inpainting , any downstream tasks.