Very happy to be back to #1 on kaggle 😁

The solution in an ensemble of segmentation (3D Unets with ResNet & B3 encoders) and object detection models (SegResNet and DynUnet backbones) from MONAI.
We accelerate model inference with TensorRT to achieve 200% speedup compared to eager PyTorch runtime

This competition challenged us to detect and classify five classes of protein complexes within a real-world cryoET dataset—pushing the boundaries of deep learning in structural biology.

Cryo-electron tomography (cryoET) generates high-resolution 3D tomograms, capturing proteins in their native, highly crowded cellular environment. It has immense potential to unlock the mysteries of the cell.

Excited to share that Eugene Khvedchenia and I secured 🏆 1st place out of 950 teams in the CryoET Object Identification Competition hosted by the Chan Zuckerberg Institute for Advanced Biological Imaging (CZII) on Kaggle! www.kaggle.com/competitions...

With 12h to go people are desperate at @kaggle.com Cryo competition

Crazy. Kaggle banned users for using YOLO 8.3.41. It was compromised and contained a crypto miner.

www.kaggle.com/discussions/...

💡 We've just released KV cache quantization in kvpress, our open source package for KV cache compression. Check it out : github.com/NVIDIA/kvpress.

Special thanks for Arthur Zucker and Marc Sun from @huggingface.bsky.social for their support 🤗

GPUs are *fast* at aligning protein sequences (and profiles!). 178x faster than JackHMMER!

In ColabFold, 23x faster end-to-end compared to AlphaFold2 reaching the same accuracy!

You get immediate speedups for all methods leveraging MSAs, from DCA, to PoET, and even AlphaFold3!

hey