Multi-instrument foundation model for large-scale heliophysics with SDO data.

Training backend for Cell Observatory models

Physiology-Augmented Chest X-Ray Masked Autoencoder

[ECCV2024] Video Foundation Models & Data for Multimodal Understanding

🌌 Applying artificial intelligence on gravitational lensing 🪐

MV-MAE is a hierarchical video model that leverages motion vectors and I-frames from compressed videos to efficiently learn masked motion representations for accurate UAV action recognition.

Masked Modeling Duo: Towards a Universal Audio Pre-training Framework

Masked Spectrogram Modeling using Masked Autoencoders for Learning General-purpose Audio Representations

[MedIA 2026] Hi-End-MAE: Hierarchical encoder-driven masked autoencoders are stronger vision learners for medical image segmentation

InfoMAE:Pair-Efficient Cross-Modal Alignment for Multimodal Time-Series Sensing

Masked Autoencoder Pretraining on 3D Brain MRI

Deep learning models for 3d volumetric ink detection on ancient Vesuvius scroll fragments.

Investigate possibilities for Vision Transformers with multiscale grids

[WACV'25] Official implementation of "PK-YOLO: Pretrained Knowledge Guided YOLO for Brain Tumor Detection in Multiplane MRI Slices".

[arXiv preprint] 🌊CascadeFormer: A Family of Two-stage Cascading Transformers for Skeleton-based Human Action Recognition

A MAE-based self-supervised setup for aortic valve detection. The model is pretrained for 400 epochs with high masking to avoid overfitting, and the resulting encoder features are used entirely within a YOLO-based pipeline for downstream valve detection.

Salient Object Detection for Video Masked Auto-Encoders

Latent Diffusion Models with Masked AutoEncoders (LDMAE) official code

[TGRS 2024] PEMAE: Pixel-Wise Ensembled Masked Autoencoder for Multispectral Pan-Sharpening

This repo reproduces key findings from Masked Autoencoders Are Scalable Vision Learners (MAE) on CIFAR-10: self-supervised pretraining improves downstream classification versus training from scratch, and we studied how decoder depth and decoder width affect MAE pretraining and downstream results.