BLIP-2 leverages frozen image encoders and large language models via a lightweight Querying Transformer, achieving state-of-the-art zero-shot performance with minimal trainable parameters.

Junnan Li, Dongxu Li, Silvio Savarese et al.

Proposes 3DShape2VecSet, a vector-set neural field for improved 3D shape encoding and diffusion-based generation.

Biao Zhang, Jiapeng Tang, Matthias Niessner et al.

Proposes end-to-end high-res diffusion with noise schedule adjustment, partial architecture scaling, and downsampling, achieving SOTA on ImageNet.

Emiel Hoogeboom, Jonathan Heek, Tim Salimans

HexPlane employs six orthogonal spatial-temporal feature planes for explicit dynamic scene representation, achieving over 100× faster training without quality loss.

Ang Cao, Justin Johnson

Improved outlier-aware segmentation using mask-level recognition, significantly reducing false positives at semantic borders.

Matej Grcić, Josip Šarić, Siniša Šegvić

Proposed the joint Caption Grounding and Generation (CGG) framework, achieving 6.8% mAP boost on novel classes in open vocabulary instance segmentation.

Jianzong Wu, Xiangtai Li, Henghui Ding et al.

Muse: a masked Transformer for text-to-image generation with FID 6.06, enabling fast, high-fidelity synthesis.

Huiwen Chang, Han Zhang, Jarred Barber et al.

Proposes Tune-A-Video, a one-shot tuning method for text-to-video generation using pretrained diffusion models, achieving high content consistency with minimal fine-tuning.

Jay Zhangjie Wu, Yixiao Ge, Xintao Wang et al.

Objaverse 1.0 is a large-scale dataset with over 800K annotated 3D models, enabling advances in 3D generative modeling, fine-grained classification, and embodied AI navigation.

Matt Deitke, Dustin Schwenk, Jordi Salvador et al.

Imagen Editor enhances text-guided image inpainting using cascaded diffusion models and EditBench benchmark.

Su Wang, Chitwan Saharia, Ceslee Montgomery et al.

ISVOS combines instance understanding with memory matching, reaching 87.1% J&F on DAVIS 2017 validation.

Junke Wang, Dongdong Chen, Zuxuan Wu et al.

BEVBert introduces spatial-aware multimodal map pre-training, boosting VLN success rates by over 4% on benchmarks.

Dong An, Yuankai Qi, Yangguang Li et al.

Hierarchical multimodal transformer Hi-VT5 processes multi-page documents, achieving 50.01% accuracy and 0.6572 ANLS on MP-DocVQA.

Rubèn Tito, Dimosthenis Karatzas, Ernest Valveny

InternVideo combines generative and discriminative self-supervised learning, achieving 91.1% top-1 accuracy on Kinetics-400 across 39 video tasks.

Yi Wang, Kunchang Li, Yizhuo Li et al.

CoupAlign couples sentence-mask and word-pixel alignments, improving oIoU by ~2% on RIS datasets.

Zicheng Zhang, Yi Zhu, Jianzhuang Liu et al.

ObjectStitch uses conditional diffusion models for label-free generative object compositing, enhancing image realism.

Yizhi Song, Zhifei Zhang, Zhe Lin et al.

Super-CLEVR isolates factors like visual complexity to enhance VQA model robustness; P-NSVQA excels in three metrics.

Zhuowan Li, Xingrui Wang, Elias Stengel-Eskin et al.

Proposes point-language association via multi-view captioning and contrastive learning, boosting open-vocabulary 3D scene understanding by 25.8%-44.7%.

Runyu Ding, Jihan Yang, Chuhui Xue et al.

OpenScene employs CLIP embedding for zero-shot 3D scene understanding, enabling open-vocabulary queries with dense point features.

Songyou Peng, Kyle Genova, Chiyu "Max" Jiang et al.

UniD3 model achieves multimodal generation via unified discrete diffusion, matching SOTA performance.

Minghui Hu, Chuanxia Zheng, Heliang Zheng et al.