Papers with Code 아카이브

Image Restoration

태스크의 하위 분류와 논문·코드·평가·기법·데이터셋을 탐색하세요.

하위 태스크
13
평가표
3
보존된 평가 기록
17

Image Restoration is a family of inverse problems for obtaining a high quality image from a corrupted input image. Corruption may occur due to the image-capture process (e.g., noise, lens blur), post-processing (e.g., JPEG compression), or photography in non-ideal conditions (e.g., haze, motion blur).

Source: Blind Image Restoration without Prior Knowledge

데이터셋

출처에 같은 태스크 이름으로 연결된 데이터셋입니다. 동명 태스크가 포함될 수 있으므로, 비교할 때는 평가 결과의 조건도 확인하세요.

Fingerprint inpainting and denoising

Synthetic training set: This set is constructed in the following two steps and will be used for estimation/training purposes. i) 84,000 275 pixel x 400 pixel ground-truth fingerprint images without any noise or scratches, but with random transformations (at most five pixels translation and +/-10 degrees rotation) were generated by using the software Anguli: Synthetic Fingerprint Generator. ii) 84,000 275 pixel x 400 pixel degraded fingerprint images were generated by applying random artifacts (blur, brightness, contrast, elastic transformation, occlusion, scratch, resolution, rotation) and backgrounds to the ground-truth fingerprint images. In total, it contains 168,000 fingerprint images (84,000 fingerprints, and two impressions - one ground-truth and one degraded - per fingerprint). Synthetic test set: This set is constructed similarly to the synthetic training set and will be used to evaluate the reconstruction performance. In total, it contains 16,800 fingerprint images (8,400 fingerprints and two impressions - one ground-truth and one degraded - per fingerprint). Since this set will be used for the purpose of evaluating the reconstruction performance, only the degraded and not the ground-truth fingerprint images will be provided to participants. Real test set: This set is constructed by systematically drawing fingerprint images with varying sizes from publicly available datasets. In total, it contains 1680 fingerprint images (140 fingerprints and 12 impressions - high-quality scans under operational conditions - per fingerprint). Description from: Fingerprint inpainting and denoising (WCCI'18, ECCV'18)

HRI

The HRI Dataset comprises a total of 3,200 image pairs. Each image pair comprises a clean background image, a depth image, a rain layer mask image, and a rainy image. It contains three scenes: lane, citystreet and japanesestreet, with image resolutions of $2048\times1024$. The lane scene contains 1,600 image pairs, consisting of images from 4 camera viewpoints, each viewpoint containing 100 images of different moments, and each moment containing 4 different intensities of rainy scenes. The citystreet scene contains 600 image pairs, consisting of images from 6 camera viewpoints, each viewpoint containing 25 images of different moments, and each moment containing 4 different intensities of rainy scenes. The japanesestreet scene contains 1,000 image pairs, consisting of images from 10 camera viewpoints, each viewpoint containing 25 images of different moments, and each moment containing 4 different intensities of rainy scenes. We split the HRI dataset into training set and test set according to camera viewpoints. For the lane scene, the training set contains images from 3 camera viewpoints, and the test set contains images from 1 camera viewpoint. For the citystreet scene, the training set contains images from 5 camera viewpoints, and the test set contains images from 1 camera viewpoint. For the japanesestreet scene, the training set contains images from 8 camera viewpoints, and the test set contains images from 2 camera viewpoints. Therefore, the training set comprises a total of 2,500 image pairs, and the test set comprises a total of 700 image pairs.

L1BSR

The Sentinel-2 satellite carries 12 CMOS detectors for the VNIR bands, with adjacent detectors having overlapping fields of view that result in overlapping regions in level-1 B (L1B) images. This dataset includes 3740 pairs of overlapping image crops extracted from two L1B products. Each crop has a height of around 400 pixels and a variable width that depends on the overlap width between detectors for RGBN bands, typically around 120-200 pixels. In addition to detector parallax, there is also cross-band parallax for each detector, resulting in shifts between bands. Pre-registration is performed for both cross-band and cross-detector parallax, with a precision of up to a few pixels (typically less than 10 pixels).

PixelShift200

Advanced pixel shift technology is employed to perform a full color sampling of the image. Pixel shift technology takes four samples of the same image at nearly the same time, and physically controls the camera sensor to move one pixel horizontally or vertically at each sampling to capture all color information at each pixel. The pixel shift technology ensures that the sampled images follow the distribution of natural images sampled by the camera, and the full information of the color (R, Gr, Gb, B channel) is completely obtained without any need of interpolation. In this way, the collected RGB images are artifacts-free, which leads to better training results for demosaicing related tasks. PixelShift200 Dataset contains 210 high quality 4K images. - Training: 200 images - Testing: 10 images - Key Features: fully colored, demosiacing artifacts free - Camera: SONY α7R III Source: PixelShift200

RawNIND

The Raw Natural Image Noise Dataset (RawNIND) is a diverse collection of paired raw images designed to support the development of denoising models that generalize across sensors, image development workflows, and styles. RawNIND contains 2831 raw images (2279 noisy and 562 clean) split across 310 scenes and 11 cameras.

SIDL: A Real-World Dataset for Restoring Smartphone Images with Dirty Lenses

Smartphone cameras are ubiquitous in daily life, yet their performance can be severely impacted by dirty lenses, leading to degraded image quality. This issue is often overlooked in image restoration research, which assumes ideal or controlled lens conditions. To address this gap, we introduced SIDL (Smartphone Images with Dirty Lenses), a novel dataset designed to restore images captured through contaminated smartphone lenses. SIDL contains diverse real-world images taken under various lighting conditions and environments. These images feature a wide range of lens contaminants, including water drops, fingerprints, and dust. Each contaminated image is paired with a clean reference image, enabling supervised learning approaches for restoration tasks. To evaluate the challenge posed by SIDL, various state-of-the-art restoration models were trained and compared on this dataset. Their performances achieved some level of restoration but did not adequately address the diverse and realistic nature of the lens contaminants in SIDL. This challenge highlights the need for more robust and adaptable image restoration techniques for restoring images with dirty lenses. Project website: https://sidl-benchmark.github.io Source : SIDL: A Real-World Dataset for Restoring Smartphone Images with Dirty Lenses

UG^2

Contains three difficult real-world scenarios: uncontrolled videos taken by UAVs and manned gliders, as well as controlled videos taken on the ground. Over 160,000 annotated frames forhundreds of ImageNet classes are available, which are used for baseline experiments that assess the impact of known and unknown image artifacts and other conditions on common deep learning-based object classification approaches. Source: UG^2: a Video Benchmark for Assessing the Impact of Image Restoration and Enhancement on Automatic Visual Recognition

FlareReal600

The FlareReal600 is a nighttime flare removal dataset, which contains 650 real-captured images pairs and 500 flare images. The training set contains 600 images pairs and 500 flare images and the validation set contains 50 image pairs. Images pairs within the dataset are captured from various place (e.g., street, park, indoor) and under incorrect & correct exposure settings. Each flare-corrupted image contains various light sources. Flare images are captured from a dark room with multiple-color light sources.

KID-F

Description K-pop Idol Dataset - Female (KID-F) is the first dataset of K-pop idol high quality face images. It consists of about 6,000 high quality face images at 512x512 resolution and identity labels for each image. We collected about 90,000 K-pop female idol images and crop the face from each image. And we classified high quality face images. As a result, there are about 6,000 high quality face images in this dataset. There are 300 test datasets for a benchmark. There are no duplicate images between test and train images. Some identities in test images are not duplicated with train images. (It means some test images is new identity to the trained model) Each test images have its degraded pair. You can use these degraded test images for testing face super resolution performance. We also provide identity labels for each image. You can download the csv file from our github # Download You can download dataset from here. Google Drive # Agreement - The use of this software is RESTRICTED to non-commercial research and educational purposes. - All images of the KID-F dataset are obtained from the internet which are not property of EDA(PCEO-AI-CLUB). EDA is not responsible for the content nor the meaning of these images. - You agree not to reproduce, duplicate, copy, sell, trade, resell or exploit for any commercial purposes, any portion of the images and any portion of derived data. - You agree not to further copy, publish or distribute any portion of the KID-F dataset. Except, for internal use at a single site within the same organization it is allowed to make copies of the dataset. - EDA reserves the right to terminate your access to the CelebA dataset at any time.

Underwater Object Detection Dataset

Description: <a href="https://gts.ai/dataset-download/underwater-object-detection-dataset/" target="_blank">👉 Download the dataset here</a> This dataset is designed for advanced underwater object detection and classification. It provides a comprehensive collection of images featuring underwater objects, each precisely annotated with bounding boxes. The dataset aims to support a wide range of research applications, from environmental monitoring to underwater robotics. Download Dataset Classes: Fish (individual and grouped) Crab Human Diver Trash (marine pollution) Jellyfish Coral Reef Sea Turtle Starfish Dataset Structure: Training Set (70%): A robust sample for building detection models. Validation Set (10%): Used to fine-tune model performance. Test Set (20%): A carefully selected set of images for evaluating model accuracy. Pre-processing Techniques: Auto-Orientation: Ensures all images are correctly aligned. Resizing: Images are scaled to 640×640 pixels for uniformity. Brightness Normalization: Corrects for underwater lighting conditions. Contrast Stretching: Enhances visibility for objects in murky or low-contrast scenes. New Annotation Techniques: Polygonal Segmentation: Introduces more precise segmentation for irregular shapes such as coral reefs. 3D Depth Mapping: For enhanced understanding of object placement in underwater space. Dataset Use Cases: Marine Ecology: Assessing species diversity and tracking the impact of environmental changes. Pollution Analysis: Detecting and classifying marine trash, aiding in cleanup efforts. Underwater Robotics: Training AUVs to recognize and navigate around complex underwater structures like coral reefs or large groups of fish. Conclusion: The expanded Underwater Object Detection provides a rich resource for researchers, environmentalists, and engineers working on underwater object detection and classification. Its enhanced classes, precise annotations, and preprocessing techniques make it a valuable asset for developing robust models in marine exploration and conservation. This dataset is sourced from Kaggle.

출처: Papers with Code · CC BY-SA 4.0. 아카이브 자료와 새로 수집한 논문은 별도로 관리합니다.