SenseNova-Vision-7B-MoT
A 7B-parameter multi-task vision-language model built on a Mixture of Transformers (MoT) architecture. It supports visual tasks including depth estimation, panoptic segmentation, pose estimation, and text-guided segmentation.
Timeline (last 90 days)
The MoT in SenseNova-Vision-7B-MoT refers to Mixture of Transformers, an architecture that combines different types of expert modules within a single model, enabling it to handle vision-language multi-task inputs and outputs.
A Reddit user, through the LibreYOLO framework, under 4-bit quantization, ran SenseNova-Vision-7B-MoT on a rented V100, using four COCO validation set photos to request four visual outputs at once: depth, panoptic segmentation, pose and text-guided segmentation.
In a complex scene containing twelve children, the depth map generated by SenseNova-Vision-7B-MoT maintained clear contours, with no obvious edge blurring or depth value confusion.
SenseNova-Vision-7B-MoT's pose estimation successfully detected the bus driver under conditions behind the windshield, at a long distance and partially occluded.
SenseNova-Vision-7B-MoT's text-guided segmentation function allows users to invoke the model's capabilities through natural language instructions, without the need to retrain a classification head for each new task.
The Reddit user's experiment was limited in scale, using only four COCO validation set photos, with no quantitative metric comparisons and no accuracy comparison with specialized models, so it is more like a feasibility verification than a comprehensive performance evaluation.
All Facts (6)
The MoT in SenseNova-Vision-7B-MoT refers to Mixture of Transformers, an architecture that combines different types of expert modules within a single model, enabling it to handle vision-language multi-task inputs and outputs.
50%UnverifiedSenseNova-Vision-7B-MoT's text-guided segmentation function allows users to invoke the model's capabilities through natural language instructions, without the need to retrain a classification head for each new task.
50%UnverifiedThe Reddit user's experiment was limited in scale, using only four COCO validation set photos, with no quantitative metric comparisons and no accuracy comparison with specialized models, so it is more like a feasibility verification than a comprehensive performance evaluation.
50%UnverifiedA Reddit user, through the LibreYOLO framework, under 4-bit quantization, ran SenseNova-Vision-7B-MoT on a rented V100, using four COCO validation set photos to request four visual outputs at once: depth, panoptic segmentation, pose and text-guided segmentation.
50%UnverifiedIn a complex scene containing twelve children, the depth map generated by SenseNova-Vision-7B-MoT maintained clear contours, with no obvious edge blurring or depth value confusion.
50%UnverifiedSenseNova-Vision-7B-MoT's pose estimation successfully detected the bus driver under conditions behind the windshield, at a long distance and partially occluded.
50%