Qiang Hu

icon indicating a website link https://huster-hq.github.io/

icon company Huazhong University of Science and Technology icon location China, Wuhan icon location a Rookie

Results 8 comments of Qiang Hu

Dice score is not change

You can try changing the optimizer

Dice coefficient no change during training,is always very close to 0

You can try changing the optimizer.

Detection and Localization

请问分割模型输出的分割掩膜是如何转化为bounding boxes的，具体使用了哪些算法？

GRPO with Qwen2.5-VL support？

> Working on it! now finished?

Motivation

Wouldn't it be advantageous if there were differences in features between lesion frames and none-lesion frames?

Transfer concept from one image to another

> This time I did not duplicate the cat, just extended the image by adding more background, you can see that the model is so stubborn to predict an object...

Transfer concept from one image to another

> , where I have mistakenly prompted a wrong bounding box for the cat on the left, [x1,y1,x2,y2] instead of [x,y,w,h], you can see that the yellow dashed bounding box...

Few-Shot Fine-tuning

> Could you let us know what questions you still have after reading https://github.com/facebookresearch/sam3/blob/main/README_TRAIN.md? Are the two datasets listed in the (https://github.com/facebookresearch/sam3/blob/main/README_TRAIN.md) both used for training SAM3 in image-level tasks?