## Radar Charts: Object and Spatial Cognition Performance ### Overview The image presents two radar charts comparing the performance of six different models (Genimi-2.5-Pro, Qwen2.5-VL-72B, VideoRefer-VL3-7B, RoboBrain-2.0-32B, RGA3-7B, and RynnEC-7B) across various cognitive tasks. The left chart focuses on "Object Cognition," while the right chart focuses on "Spatial Cognition." Each axis represents a different task or attribute, and the distance from the center indicates the model's performance on that task. ### Components/Axes **Legend (Top-Left):** * **Orange:** Genimi-2.5-Pro * **Green:** Qwen2.5-VL-72B * **Blue:** VideoRefer-VL3-7B * **Red:** RoboBrain-2.0-32B * **Gray:** RGA3-7B * **Purple:** RynnEC-7B (Ours) **Left Chart - (a) Object Cognition:** * **Axes (Clockwise from Top):** Category, Color, Material, Shape, State, Position, Function, Surface, Size, Situational Seg, Direct Seg, Counting * **Scale:** The scale is implied to be from 0 to 100, based on the data values. **Right Chart - (b) Spatial Cognition:** * **Axes (Clockwise from Top):** Trajectory Review, Egocentric Direction, Egocentric Distance, Movement Imagery, Spatial Imagery, Object Height, Object Size, Object Distance, Absolute Position, Relative Position * **Scale:** The scale is implied to be from 0 to 100, based on the data values. ### Detailed Analysis **Left Chart - (a) Object Cognition:** * **Category:** * Genimi-2.5-Pro (Orange): ~28 * Qwen2.5-VL-72B (Green): ~69 * VideoRefer-VL3-7B (Blue): ~60 * RoboBrain-2.0-32B (Red): ~15 * RGA3-7B (Gray): ~11 * RynnEC-7B (Purple): ~50 * **Color:** * Genimi-2.5-Pro (Orange): ~22 * Qwen2.5-VL-72B (Green): ~60 * VideoRefer-VL3-7B (Blue): ~63 * RoboBrain-2.0-32B (Red): ~0 * RGA3-7B (Gray): ~13 * RynnEC-7B (Purple): ~50 * **Material:** * Genimi-2.5-Pro (Orange): ~14 * Qwen2.5-VL-72B (Green): ~66 * VideoRefer-VL3-7B (Blue): ~68 * RoboBrain-2.0-32B (Red): ~24 * RGA3-7B (Gray): ~7 * RynnEC-7B (Purple): ~56 * **Shape:** * Genimi-2.5-Pro (Orange): ~16 * Qwen2.5-VL-72B (Green): ~60 * VideoRefer-VL3-7B (Blue): ~60 * RoboBrain-2.0-32B (Red): ~36 * RGA3-7B (Gray): ~22 * RynnEC-7B (Purple): ~66 * **State:** * Genimi-2.5-Pro (Orange): ~3 * Qwen2.5-VL-72B (Green): ~50 * VideoRefer-VL3-7B (Blue): ~66 * RoboBrain-2.0-32B (Red): ~50 * RGA3-7B (Gray): ~33 * RynnEC-7B (Purple): ~66 * **Position:** * Genimi-2.5-Pro (Orange): ~6 * Qwen2.5-VL-72B (Green): ~66 * VideoRefer-VL3-7B (Blue): ~70 * RoboBrain-2.0-32B (Red): ~46 * RGA3-7B (Gray): ~25 * RynnEC-7B (Purple): ~60 * **Function:** * Genimi-2.5-Pro (Orange): ~44 * Qwen2.5-VL-72B (Green): ~63 * VideoRefer-VL3-7B (Blue): ~60 * RoboBrain-2.0-32B (Red): ~30 * RGA3-7B (Gray): ~54 * RynnEC-7B (Purple): ~63 * **Surface:** * Genimi-2.5-Pro (Orange): ~77 * Qwen2.5-VL-72B (Green): ~49 * VideoRefer-VL3-7B (Blue): ~63 * RoboBrain-2.0-32B (Red): ~25 * RGA3-7B (Gray): ~75 * RynnEC-7B (Purple): ~63 * **Size:** * Genimi-2.5-Pro (Orange): ~77 * Qwen2.5-VL-72B (Green): ~30 * VideoRefer-VL3-7B (Blue): ~46 * RoboBrain-2.0-32B (Red): ~30 * RGA3-7B (Gray): ~75 * RynnEC-7B (Purple): ~46 * **Situational Seg:** * Genimi-2.5-Pro (Orange): ~28 * Qwen2.5-VL-72B (Green): ~56 * VideoRefer-VL3-7B (Blue): ~36 * RoboBrain-2.0-32B (Red): ~28 * RGA3-7B (Gray): ~44 * RynnEC-7B (Purple): ~36 * **Direct Seg:** * Genimi-2.5-Pro (Orange): ~45 * Qwen2.5-VL-72B (Green): ~60 * VideoRefer-VL3-7B (Blue): ~23.5 * RoboBrain-2.0-32B (Red): ~82.8 * RGA3-7B (Gray): ~54 * RynnEC-7B (Purple): ~69 * **Counting:** * Genimi-2.5-Pro (Orange): ~28 * Qwen2.5-VL-72B (Green): ~69 * VideoRefer-VL3-7B (Blue): ~15 * RoboBrain-2.0-32B (Red): ~11 * RGA3-7B (Gray): ~44 * RynnEC-7B (Purple): ~69 **Right Chart - (b) Spatial Cognition:** * **Trajectory Review:** * Genimi-2.5-Pro (Orange): ~30 * Qwen2.5-VL-72B (Green): ~97 * VideoRefer-VL3-7B (Blue): ~41 * RoboBrain-2.0-32B (Red): ~15 * RGA3-7B (Gray): ~29 * RynnEC-7B (Purple): ~90 * **Egocentric Direction:** * Genimi-2.5-Pro (Orange): ~22 * Qwen2.5-VL-72B (Green): ~77 * VideoRefer-VL3-7B (Blue): ~31 * RoboBrain-2.0-32B (Red): ~21 * RGA3-7B (Gray): ~9 * RynnEC-7B (Purple): ~41 * **Egocentric Distance:** * Genimi-2.5-Pro (Orange): ~27 * Qwen2.5-VL-72B (Green): ~40 * VideoRefer-VL3-7B (Blue): ~46 * RoboBrain-2.0-32B (Red): ~34 * RGA3-7B (Gray): ~12 * RynnEC-7B (Purple): ~5 * **Movement Imagery:** * Genimi-2.5-Pro (Orange): ~6 * Qwen2.5-VL-72B (Green): ~25 * VideoRefer-VL3-7B (Blue): ~59 * RoboBrain-2.0-32B (Red): ~36 * RGA3-7B (Gray): ~3 * RynnEC-7B (Purple): ~15 * **Spatial Imagery:** * Genimi-2.5-Pro (Orange): ~37 * Qwen2.5-VL-72B (Green): ~30 * VideoRefer-VL3-7B (Blue): ~15 * RoboBrain-2.0-32B (Red): ~8 * RGA3-7B (Gray): ~0 * RynnEC-7B (Purple): ~48 * **Object Height:** * Genimi-2.5-Pro (Orange): ~47 * Qwen2.5-VL-72B (Green): ~39 * VideoRefer-VL3-7B (Blue): ~15 * RoboBrain-2.0-32B (Red): ~6 * RGA3-7B (Gray): ~0 * RynnEC-7B (Purple): ~21 * **Object Size:** * Genimi-2.5-Pro (Orange): ~67 * Qwen2.5-VL-72B (Green): ~66 * VideoRefer-VL3-7B (Blue): ~59 * RoboBrain-2.0-32B (Red): ~36 * RGA3-7B (Gray): ~3 * RynnEC-7B (Purple): ~48 * **Object Distance:** * Genimi-2.5-Pro (Orange): ~39 * Qwen2.5-VL-72B (Green): ~66 * VideoRefer-VL3-7B (Blue): ~46 * RoboBrain-2.0-32B (Red): ~34 * RGA3-7B (Gray): ~9 * RynnEC-7B (Purple): ~15 * **Absolute Position:** * Genimi-2.5-Pro (Orange): ~66 * Qwen2.5-VL-72B (Green): ~93 * VideoRefer-VL3-7B (Blue): ~41 * RoboBrain-2.0-32B (Red): ~31 * RGA3-7B (Gray): ~15 * RynnEC-7B (Purple): ~59 * **Relative Position:** * Genimi-2.5-Pro (Orange): ~93 * Qwen2.5-VL-72B (Green): ~97 * VideoRefer-VL3-7B (Blue): ~41 * RoboBrain-2.0-32B (Red): ~31 * RGA3-7B (Gray): ~15 * RynnEC-7B (Purple): ~90 ### Key Observations * **Qwen2.5-VL-72B (Green):** Generally performs well across both Object and Spatial Cognition tasks, often achieving the highest scores. * **Genimi-2.5-Pro (Orange):** Shows variable performance, excelling in some Object Cognition tasks (e.g., Size, Surface) but lagging in others. It also shows good performance in Relative Position. * **VideoRefer-VL3-7B (Blue):** Performance is generally moderate, with some strengths in specific areas. * **RoboBrain-2.0-32B (Red):** Tends to have lower scores across most tasks, indicating weaker performance compared to other models. * **RGA3-7B (Gray):** Generally exhibits low performance, particularly in Spatial Cognition. * **RynnEC-7B (Purple):** Shows competitive performance, especially in Object Cognition, and performs well in Trajectory Review and Relative Position in Spatial Cognition. ### Interpretation The radar charts provide a comparative analysis of the cognitive abilities of different models. Qwen2.5-VL-72B appears to be a strong all-around performer, while RoboBrain-2.0-32B and RGA3-7B generally underperform compared to the others. Genimi-2.5-Pro and RynnEC-7B show more specialized strengths. The data suggests that different models have varying proficiencies in different cognitive tasks, highlighting the importance of selecting the appropriate model based on the specific application. The "Ours" designation for RynnEC-7B suggests that this model is the focus of the study, and its performance is being compared against existing models.