temporal-attention

This work considers combine multi-tricks with highway network to achieve traffic flow prediction accurately.

🎯 Production-ready implementation of video prediction models using PyTorch. Features Enhanced ConvLSTM with temporal attention, PredRNN with spatiotemporal memory, and Transformer-based architecture.

[BMVC 2023 ] Official code for "STEP CATFormer: Spatial-Temporal Effective Body-Part Cross Attention Transformer for Skeleton-based Action Recognition"

Dynamic Knowledge Graph with Temporal Logical Attention Network for cloud-native anomaly detection and root-cause localization

Official code for FAME(Elsevier ESWA-2025): A lightweight spatio-temporal framework for Deepfake Model Attribution using VGG-based feature extraction and temporal attention. Outperforms prior methods across DFDM, FF++, and FakeAVCeleb datasets.

PyTorch U-TAE model for binary water and land segmentation from Sentinel-1 and Sentinel-2 imagery using the IBM Granite flood dataset.

Deep Reinforced Model for Abstractive Summarization

Topic-Aware Abstractive Summarization

Cell-segmentation using time-sequence data, through the adaptation of a U-TAE Model.

From-scratch flow-matching video diffusion model — factorized spatiotemporal DiT, velocity target, trained on Moving MNIST.

Official code for CapST (ACM TOMM 2025): A Capsule Network and Temporal Attention-based framework for Deepfake Model Attribution. Achieves efficient and accurate source identification across DFDM and GAN-generated datasets.

Predictive maintenance and remaining useful life forecasting using stacked GRU autoencoders, temporal attention, and NASA C-MAPSS turbofan sensor data.

Heterogeneous GAT with temporal attention for drug-drug interaction prediction. Multi-task learning (interaction + metabolite pathway + temporal dynamics) on Tox21 molecular data. 99.9% interaction accuracy, 0.875 temporal correlation.

AQuaMoHo is a ML-driven framework to annotate air quality at any location from a low-cost Thermo-Hygrometer

A lightweight audio turn-detection model for conversational AI, built with OpenAI Whisper embeddings and temporal attention. Compares acoustic and pretrained speech representations, evaluates multiple pooling strategies, and includes inference, streaming, testing, and experimental fine-tuning pipelines.