Google has introduced MagicLens, a state-of-the-art image retrieval model that outperforms previous models while being 50 times smaller. The model supports various search intents and retrieval types, achieving top results. Another project, ObjectDrop, by Google Research, enables photorealistic object removal and insertion with consideration of scene effects.
[CV] MagicLens: Self-Supervised Image Retrieval with Open-Ended Instructions K Zhang, Y Luan, H Hu, K Lee, S Qiao, W Chen, Y Su, M Chang [Google DeepMind & The Ohio State University] (2024) https://t.co/oM5hjf6FH9 - The paper introduces MagicLens, self-supervised image retrieval… https://t.co/JAtVMbatRj
Google presents ObjectDrop! The method enables photorealistic object removal and insertion while considering their effects on the scene. More examples ⬇️ https://t.co/Ir26h01OtH
[CV] ObjectDrop: Bootstrapping Counterfactuals for Photorealistic Object Removal and Insertion D Winter, M Cohen, S Fruchter, Y Pritch, A Rav-Acha, Y Hoshen [Google Research] (2024) https://t.co/0beUdHxBqU - The paper proposes a practical approach to train a diffusion model on a… https://t.co/ZnKRAF4Dhv
Thanks @arankomatsuzaki sharing our work! MagicLens support 1⃣ Open-ended search intents: simple, complex, and beyond visual ones; 2⃣ Various retrieval: text-to-image, image-to-image, and multimodal-to-image. All with #SOTA results! Check out: https://t.co/IBvpbuaBfA https://t.co/bF3UGRDLOZ
Google presents MagicLens: image retrieval models following open-ended instructions Outperforms previous SotA but with a 50x smaller model size proj: https://t.co/R01mPuvkvy abs: https://t.co/HTcp25ZxdY https://t.co/N0swbIdj17
🔍MagicLens: State-of-the-art instruction-following image retrieval model on 10 benchmarks but 50x smaller than prior best! Check out our paper on huggingface🤗: https://t.co/IkIeJ2Spx6 https://t.co/uP2Gv6XvHZ