YZ Sözlüğü
Yapay Zekanın tam sözlüğü
Low-Light Image Enhancement
Preprocessing that improves the visibility of objects in low-light scenes by amplifying the light signal while minimizing noise, often via Generative Adversarial Networks (GANs).
Weather-Resilient Detection
The ability of a system to maintain stable detection performance despite the presence of weather disturbances such as rain, snow, or fog, often via specific denoising architectures.
Part-Based Models
A detection approach that models an object as a collection of connected spatial parts, allowing for better robustness to occlusions by detecting visible sub-components even if the overall object is masked.
Multi-Scale Feature Fusion
A method combining visual features at different spatial resolutions to improve the detection of objects of varying sizes and in low-contrast conditions where details are difficult to discern.
Synthetic Data Augmentation
The generation of artificial training images simulating difficult conditions (rain, night, fog) to enrich the dataset and improve the model's robustness without having to collect corresponding real images.
Adversarial Training for Robustness
A training process where the model is confronted with examples maliciously perturbed or by difficult conditions, forcing it to learn more invariant and resistant representations.
Thermal Imaging Fusion
The integration of data from thermal infrared sensors with visible images to enable the detection of objects in total darkness or through opaque weather conditions.
Self-Supervised Learning for Robustness
A learning paradigm where the model generates its own labels from unannotated data in difficult conditions, allowing it to learn robust features without explicit supervision.
Attention Mechanisms for Occlusion
Using attention layers that allow the model to selectively focus on the visible parts of an occluded object and to weigh their importance for the final prediction.
Curriculum Learning for Adverse Conditions
A training strategy that progressively exposes the model to increasingly difficult conditions, simulating a step-by-step learning process to improve its final robustness.
Dehazing and Deraining Networks
Neural networks specialized in removing visual artifacts such as fog or rain before the detection step, acting as a preprocessing step to restore scene clarity.
Feature Pyramid Networks (FPN) for Small Objects
An architecture that builds a high-resolution, multi-scale feature pyramid, essential for detecting small objects or partially visible objects in low-quality images.
Temporal Consistency Modeling
Integrating temporal information from video sequences to reinforce detections in individual degraded images, by exploiting the consistency of objects across frames.