Enhancing Visual Grounding for GUI Agents via Self-Evolutionary Reinforcement Learning

Benchmark Model Rank Results
natural-language-visual-grounding-on-screenspotSE-GUI-7B (extra data)#11Accuracy (%): 88.2
computer-use-agents-on-screenspot-proSE-GUI-7B (extra data)#35Accuracy: 47.3
computer-use-agents-on-screenspot-v2SE-GUI-7B (extra data)#35Accuracy: 90.3