Need Visual Grounding demo with lmdeploy
Need Visual Grounding demo with lmdeploy。I used lmstudio-community/InternVL3_5-30B-A3B-GGUF, but it didn’t perform well in Visual Grounding. I noticed that its reasoning process contains many uncertainties—for instance, it is unaware of the image size, so it has to make assumptions about the image size first, and then guess the relative positions of various elements in the image. However, this approach fails to accurately frame the target objects.
Thank you for your interest in our work. Our model has not been trained on grounding tasks under the thinking mode. For such tasks, we recommend that you do not enable the thinking mode, as these tasks are typically more perception-oriented and place relatively low demands on the model’s reasoning ability.