Iris-Z: Scaling One Generative Model Past Specialist Geometry Models
One set of weights generates images with depth, estimates depth from photographs and paints images onto supplied depth.
Iris-Z uses a single backbone to process appearance and geometry together, without specialized layers. The authors report that, at sufficient scale, it beats state-of-the-art specialist models on established depth benchmarks. It also retains the image-generation abilities of its starting backbone. They are releasing two models.
HF Daily Papers' note
Iris-Z uses a single backbone to process appearance and geometry together, without specialized layers. The authors report that, at sufficient scale, it beats state-of-the-art specialist models on established depth benchmarks. It also retains the image-generation abilities of its starting backbone. They are releasing two models.
HF Daily Papers' note
score 5