Research map / Image generation / Image restoration
Image Classification in the Dark using Quanta Image Sensors
Abhiram Gnanasambandam, Stanley H. Chan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add student-teacher (perceptual loss) training instead of plain fine-tuning (ablation) | VGG-16 classifier on QIS · Dog dataset, synthetic QIS, 0.25 ppp | accuracy (%): 42.1% -> 48.6% | Improved |
| add student-teacher training instead of plain fine-tuning (ablation) | ResNet50 classifier on QIS · Dog dataset, synthetic QIS, 0.25 ppp | accuracy (%): 43.3% -> 49.1% | Improved |
| add student-teacher training instead of plain fine-tuning (ablation) | InceptionV3 classifier on QIS · Dog dataset, synthetic QIS, 0.25 ppp | accuracy (%): 44.3% -> 50.0% | Improved |
| add student-teacher learning over end-to-end Dirty Pixels training (shallow entrance) (ablation) | QIS classifier · Synthetic QIS, 1 ppp | accuracy (%): 53.9% -> 62.7% | Improved |
| retrain the classifier on low-light data instead of using a pre-trained clean classifier (Restoration Network) (ablation) | QIS classifier · Synthetic QIS, 0.25 ppp | accuracy (%): 35.6% -> 52.1% | Improved |
| use QIS sensor instead of CIS with the same classifier (ablation) | proposed classifier · Dog dataset, synthetic, 4 ppp | accuracy (%): 69.8% -> 72.9% | Improved |
| use QIS sensor instead of CIS with the same classifier (ablation) | proposed classifier · Dog dataset, synthetic, 0.25 ppp | accuracy gap (%): 3.1% -> 23.1% | Improved |
| use a 20-layer UNet deep denoiser instead of a 2-layer shallow entrance with student-teacher learning (ablation) | proposed method on QIS · Synthetic QIS | accuracy | No clear effect |
| use a deep denoiser instead of shallow entrance (ablation) | Dirty Pixels on QIS · Synthetic QIS | accuracy | Improved |
| use a deep denoiser on CIS data (ablation) | classifier on CIS · Synthetic CIS, signal well below read noise | accuracy | Improved |
Closest papers by meaning
- Dynamic Low-light Imaging with Quanta Image Sensors (ECCV 2020 · similarity 0.90)
- Quanta Neural Networks: From Photons to Perception (ICCV 2025 · similarity 0.88)
- Streaming Quanta Sensors for Online, High-Performance Imaging and Vision (TPAMI 2024 · similarity 0.85)
- Seeing Into Darkness: Scotopic Visual Recognition (CVPR 2017 · similarity 0.84)
- Photon-Starved Scene Inference Using Single Photon Cameras (ICCV 2021 · similarity 0.83)
- bit2bit: 1-bit quanta video reconstruction via self-supervised photon prediction (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
