Adaptive Camera Sensor for Vision Models

Authors: Eunsu Baek, Sung-hwan Han, Taesik Gong, Hyung-Sin Kim

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on both Image Net-ES and our new Image Net-ES Diverse show that Lens significantly improves model accuracy across various baseline schemes for sensor control and model modification, while maintaining low latency in image captures. Lens effectively compensates for large model size differences and integrates synergistically with model improvement techniques.
Researcher Affiliation Academia Eunsu Baek* Graduate School of Data Science Seoul National University Seoul, Republic of Korea EMAIL Sung-Hwan Han Department of Computer Science & ngineering Seogang University Seoul, Republic of Korea sunghwan.Na EMAIL Taesik Gong Department of Computer Science & Engineering Ulsan National Institute of Science and Technology Ulsan, Republic of Korea EMAIL Hyung-Sin Kim Graduate School of Data Science Seoul National University Seoul, Republic of Korea EMAIL
Pseudocode No The paper includes diagrams and describes algorithms like CSAs in paragraph text, but does not contain a clearly labeled pseudocode block or algorithm section with structured steps formatted like code.
Open Source Code Yes Our code and dataset are available at github.com/Edw2n/Lens.git.
Open Datasets Yes To validate Lens, we introduce Image Net-ES Diverse, a new benchmark dataset capturing natural perturbations from varying sensor and lighting conditions. [...] Our code and dataset are available at github.com/Edw2n/Lens.git. [...] We release Image Net-ES Diverse, a new benchmark dataset containing 192,000 images that capture natural covariate shifts through varying sensor and lighting conditions.
Dataset Splits Yes Table 9: Environment and Sensor specifics of Image Net-ES (Baek et al., 2024). Dataset Test Original samples 1,000 [...] Validation Original samples 1,000. [...] Five images were randomly selected from each of the 200 classes in the Tiny-Image Net validation set (Le & Yang, 2015).
Hardware Specification Yes Table 6: Comparison on Testbed between Luminous and Diverse. Comparison Components ES-Studio Diverse Specifications Control system Desktop Computer + Wifi Network Apple Mac Studio M2 Max + CCAPI + Philips Hue API
Software Dependencies No The paper mentions using libraries like 'timm library (Wightman, 2019)' and 'Torchvision Marcel & Rodriguez (2010) library' and 'Open OOD (Yang et al., 2022)', but does not provide specific version numbers for these software dependencies within the main text or appendices.
Experiment Setup Yes Table 12: Description of underlying models for OOD detection experiments. (Optimizer: SGD, Scheduler: Reduce LROn Plateau, Batch size: 128) Model ... Training configuration Efficient Net-B0 ... lr: 5e-3, epochs: 20 Res Net18 ... lr: 5e-2, epochs: 15 Dei T ... lr: 5e-3 , epochs: 20 Swin-B ... lr: 5e-3 , epochs: 20