FunAI Lab's Open Vision Model Beats Google's Vision Banana on Cityscapes Benchmark

FunAI Lab published the training code and weights for a computer vision finetune of the FLUX.2 Klein 4B and 9B image generators. The new model achieved a 73.9 mIoU score on the Cityscapes benchmark, surpassing Google DeepMind's Vision Banana's 69.9 mIoU. The project was developed using only public data and released 148 checkpoints for the community.