Collection of best papers from top AI conferences (github.com)

This repository is a curated collection of standout papers from top AI/ML conferences (CVPR, NeurIPS, ICLR, ICCV, ECCV, ICML, AAAI, WACV, BMVC) spanning foundational classics (ImageNet, FCN, Inception, DenseNet, StyleGAN) to bleeding‑edge 2025 work. Recent highlights include CVPR 2025 contributions such as VGGT (Visual Geometry Grounded Transformer), MegaSaM (fast, accurate structure & motion from casual dynamic videos), Navigation World Models (LeCun et al.) and Molmo/PixMo — which explicitly emphasize open weights and open data for state‑of‑the‑art vision‑language models. The list also collects modern advances in neural rendering and 3D reconstruction (pixelSplat, Mip‑Splatting, Neural Inverse Rendering), generative multimodal pretraining (discrete diffusion timestep tokens), and a long tail of influential papers across perception, representation learning, and embodied AI. For researchers and practitioners this is a compact lens into the field’s technical trajectory: it links test‑of‑time classics to current trends (open models/data, geometry‑aware transformers, scalable neural splatting, robust SfM from in-the-wild video, and improved multimodal pretraining). The curated scope makes it easy to trace methods, datasets and reproducible code paths that shaped current SOTA, accelerating literature review, replication, and idea generation for further research.