It would take 375,000 Lego bricks stacked one on top of another to destroy the bottom brick. The tower would be 12,000 feet tall, applying a force of 4,200N (equivalent to 950lbs). Everything has a breaking point.

TorchVision v0.9 comes with a series of new mobile-friendly models that can be used for Classification, Object Detection and Semantic Segmentation. This article looks into the code of the models and shares implementation, training, configuration and more. pytorch.org/blog/torchvisi…

What actually happens when you call .backwards() in PyTorch? Autograd goodness 🪄! PyTorch keeps track of all of the computations you’ve done on each of your tensors and .backwards() triggers it to compute the gradients and stores them in .grad. 1/3

#Albumentations 1.0.0 has been released! New version contains: - 10 new transforms - bug fixes etc See the release notes for details github.com/albumentations… Thanks a lot to core team Vladimir Iglovikov creaf Одна інтеллектуальна одиниця Alexander and everyone who helps improve the library

Check out our work on line segment detection. We present a light-weight and real-time LSD, named Mobile-LSD (M-LSD), which is available on GPU, CPU, and even on mobile devices! We also include box detection demo using M-LSD that can be used in real-world applications.

🤔 Remember when you requested a UI for DVC and CML? The day has come! 🎉 😍 UI for ML experiment tracking 💪🏼 The power of Git to track, preserve and reproduce experiments 📈 Graphs to compare experiments 🚀 Run CML in the cloud or Kubernetes 👉🏽 dvc.org/blog/introduci…

You could use Aphex or you could just use 1Password

Our mobile line segment detector (M-LSD) seems to generalize to anime and sketch figures pretty well! Try it out with Gradio web demo :) Paper: arxiv.org/abs/2106.00186 Github: github.com/navervision/ml… Gradio demo: gradio.app/g/AK391/mlsd (thanks to AK moved to @_akhaliq )

Team Gradio created GUIs that can demo outputs/results of 9 models on PyTorch Hub: - Silero STT - WaveGlow - MiDaS - YOLO v5 - X3D - NTSNET - Transformer NMT - Silero Language Identifier - PGAN View them all here: pytorch.org/hub/research-m…

Dockerfile now supports multi-line RUN with heredocs (<<eot) github.com/moby/buildkit/… --- # syntax = docker/dockerfile-upstream:master-labs FROM debian RUN <<eot apt-get update apt-get install -y vim eot

Version 1.0.0 of the most advanced template for Python projects on GitHub has been released! New generation parameters, updated commands in Makefile, dependencies refactoring. All this happened thanks to feedback from users. Check it now 📈 github.com/TezRomacH/pyth…

🎉 Announcing a new blog post 🎊 Writing agnostic distributed code can be tedious. Find out how PyTorch-Ignite makes data distributed training easy with minimal code change compared to PyTorch DDP, Horovod and XLA - labs.quansight.org/blog/2021/06/d…

dokai:21.07 - new release of docker images for DL and video processing * CUDA 11.3.1, cuDNN 8.2.0 * FFmpeg 4.4 with NVDEC, NVENC * PyTorch 1.9.0, torchvision 0.10.0 * timm 0.4.12 github.com/osai-ai/dokai

#Albumentations 1.1.0 is out! The new release of a fast and flexible library for image augmentation includes new transformations and improvements for the current ones. Thanks to Mikhail Druzhinin Vladimir Iglovikov Одна інтеллектуальна одиниця Alexander and all our contributors. github.com/albumentations…

1st place solution for the SoccerNet Ball Action Spotting Challenge 2023 at the CVSports #CVPR2023 - Efficient model for video action detection - Multi-stage training (transfer learning, sampling techniques) - Fast GPU-based video loading for training github.com/lRomul/ball-ac…

SoccerNet Ball Action Spotting というCVPR Sports WS内のコンペに筑波、東大、名大の学生と参加して4位でした(悔しい) 放送映像からボールタッチ{Pass, Drive}を検出するタスクで，可視化結果を見ると割と検出出来てそうですが...中々厳しい 右上が予測ラベルで左下がconfidence #CVPR #SoccerNet

The Sensorium 2023 competition is over. Here are the winning teams: 🥇IRomul 🥈YuZhu 🥉dunedin All winners used custom architectures. To learn more about their approaches, join our in-person workshop at NeurIPS on Friday, Dec 15, 2023. 🗓️ 1/4

NeurIPS Sensorium Competition workshop is happening on Friday from 1:30–4:30 in room 357. Explore with us the state of the art and future of predictive models of neural responses. Keynotes by Daniel Yamins, Colin Conwell and SueYeon Chung

DwiseNeuro is a novel architecture for predicting V1 neural activity in mice in response to videos. The architecture was part of the winning solution for the Sensorium at NeurIPS 2023. github.com/lRomul/sensori…