mlsanityData to 7 Oct 2026

MotionDeltaCNN: Sparse CNN Inference of Frame Differences in Moving Camera Videos

arXiv 2210.09887 ↗First posted 18 Oct 2022cs.CV cs.LGcited by 5

Not on the map. Why?

Abstract (arXiv)

Convolutional neural network inference on video input is computationally expensive and requires high memory bandwidth. Recently, DeltaCNN managed to reduce the cost by only processing pixels with significant updates over the previous frame. However, DeltaCNN relies on static camera input. Moving cameras add new challenges in how to fuse newly unveiled image regions with already processed regions efficiently to minimize the update rate - without increasing memory overhead and without knowing the camera extrinsics of future frames. In this work, we propose MotionDeltaCNN, a sparse CNN inference framework that supports moving cameras. We introduce spherical buffers and padded convolutions to enable seamless fusion of newly unveiled regions and previously processed regions -- without increasing memory footprint. Our evaluation shows that we outperform DeltaCNN by up to 90% for moving camera videos.

Key facts. Cited by 5 papers in the map (data through 7 Oct 2026). Builds on 24 of its 40 references that are in the map. First posted on arXiv 18 Oct 2022.

Builds on

24 of 40 references.

References · 24

  1. 2016Densely Connected Convolutional Networkscited by 6,691
  2. 2019EfficientNet: Rethinking Model Scaling for Convolutional Neural Networkscited by 6,415
  3. 2015Learning both Weights and Connections for Efficient Neural Networkscited by 1,998
  4. 2019Deep High-Resolution Representation Learning for Human Pose Estimationcited by 1,339
  5. 2017The 2017 DAVIS Challenge on Video Object Segmentationcited by 942
  6. 20173D Semantic Segmentation with Submanifold Sparse Convolutional Networkscited by 746
  7. 2018Simple Baselines for Human Pose Estimation and Trackingcited by 502
  8. 2016Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activationscited by 413
  9. 2016EIE: Efficient Inference Engine on Compressed Deep Neural Networkcited by 291
  10. 2016Deep Feature Flow for Video Recognitioncited by 186
  11. 2015Fixed Point Quantization of Deep Convolutional Networkscited by 141
  12. 2018Accel: A Corrective Fusion Network for Efficient Semantic Segmentation on Videocited by 78
12 more
  1. 2018Low-Latency Video Semantic Segmentationcited by 61
  2. 2017Towards High Performance Video Object Detectioncited by 60
  3. 2016Clockwork Convnets for Video Semantic Segmentationcited by 55
  4. 2021Skip-Convolutions for Efficient Video Processingcited by 32
  5. 2019Dynamic Kernel Distillation for Efficient Pose Estimation in Videoscited by 30
  6. 2017Minimum Energy Quantized Neural Networkscited by 20
  7. 2022DeltaCNN: End-to-End CNN Inference of Sparse Frame Differences in Videoscited by 18
  8. 2018Recurrent Residual Module for Fast Inference in Videoscited by 13
  9. 2021Pixel-Level Bijective Matching for Video Object Segmentationcited by 11
  10. 2022Delta Distillation for Efficient Video Processingcited by 10
  11. 2018CBinfer: Exploiting Frame-to-Frame Locality for Faster Convolutional Network Inference on Video Streamscited by 6
  12. 2021Event Neural Networkscited by 5

Built on by

Cited by year

Citing papers per year: 2024: 2; 2025: 3.

5 papers.

  1. 2024MaskVD: Region Masking for Efficient Video Object Detectioncited by 5
  2. 2025D'ej`a Vu: Efficient Video-Language Query Engine with Learning-based Inter-Frame Computation Reusecited by 5
  3. 2025DeltaLLM: A Training-Free Framework Exploiting Temporal Sparsity for Efficient Edge LLM Inferencecited by 4
  4. 2024Energy-Efficient & Real-Time Computer Vision with Intelligent Skipping via Reconfigurable CMOS Image Sensorscited by 1
  5. 2025ReFrame: Layer Caching for Accelerated Inference in Real-Time Renderingcited by 0