Pinned
Today we are launching Marble – a multimodal world model that lets you create and edit 3D worlds.
Introducing Marble by World Labs: a foundation for a spatially intelligent future.
Create your world at marble.worldlabs.ai
00:00
Justin Johnson
394 posts
Justin Johnson
@jcjohnss
Joined January 2014
- Videos for my Fall 2019 course "Deep Learning for Computer Vision" are now on YouTube! This is an evolution of @cs231n that I used to teach at Stanford: - All content refreshed - New topics: Transformers, Video, 3D, etc - HW in @PyTorch + @GoogleColab
- I'm excited about Segment Anything released from FAIR today. It tackles an old problem (find objects in images) at large scale: trained on 11M images and 1B objects. This is a new Foundation Model for Computer Vision - it recognizes any object in any context.
GIF - My new paper on generating images from scene graphs using graph convolution and GANs is up on arXiv! To appear at CVPR2018, with @agrimgupta92 and @drfeifei arxiv.org/abs/1804.01622
- Our new paper (w/@kdexd) argues that "language is all you need" for good visual features: we train CNN+Transformer *from scratch* on ~100k images+captions from COCO, transfer the CNN to 6 downstream vision tasks, and match/exceed ImageNet features despite using 10x fewer images!Introducing "VirTex": a pretraining approach to learn visual features via language using fewer images. Pretrain: CNN+Transformer from scratch on COCO Captions. Transfer CNN: Results on 6 vision tasks match/exceed ImageNet pretraining (10x size wrt COCO)! arxiv.org/abs/2006.06666
- Today we released PyTorch3D v0.2, adding new features around point clouds: - Point cloud renderer - Point-to-mesh distances - Normal estimation - Umeyama, ICP, PnP, and KNN All batched and differentiable, ready to drop into your deep learning models!
- Today we released code for SynSin, our CVPR'20 oral that generates novel views from a single image: github.com/facebookresear… We have: - Pretrained models - Jupyter notebook demos - Training and evaluation - #pytorch3d integration Congrats to @oliviawiles1 on the release!
GIF - I successfully defended my PhD today! Huge thanks to @drfeifei for being an amazing mentor on this journey!
- This week we open-sourced pycls, a flexible research framework for image classification with @PyTorch encapsulating current best practices. Used internally for research @FacebookAI -- excited to share with the community! Led by Ilija Radosavovic @ir413
- PyTorch3D is our new library for accelerating 3D deep learning research, and provides: - Easy batching of heterogeneous triangle meshes - Optimized implementations of common mesh ops - Modular, efficient, differentiable mesh renderer - More to come!We just released PyTorch3D, a new toolkit for researchers and engineers that’s fast and modular for 3D deep learning research: ai.facebook.com/blog/-introduc…
00:00 - Project reports for #cs231n 2017 are now online! 250+ amazing applications of deep learning: cs231n.stanford.edu/reports.html
- I'm excited to share our new paper that jointly detects objects and predicts 3D triangle meshes in real-world images, called Mesh R-CNN. With Georgia Gkioxari and Jitendra Malik arxiv.org/abs/1906.02739
