PK PyTorchKR Papers
Papers with Code 아카이브

Image Matting

태스크의 하위 분류와 논문·코드·평가·기법·데이터셋을 탐색하세요.

전체 기법전체 데이터셋
하위 태스크
1
평가표
8
보존된 평가 기록
45

Image Matting is the process of accurately estimating the foreground object in images and videos. It is a very important technique in image and video editing applications, particularly in film production for creating visual effects. In case of image segmentation, we segment the image into foreground and background by labeling the pixels. Image segmentation generates a binary image, in which a pixel either belongs to foreground or background. However, Image Matting is different from the image segmentation, wherein some pixels may belong to foreground as well as background, such pixels are called partial or mixed pixels. In order to fully separate the foreground from the background in an image, accurate estimation of the alpha values for partial or mixed pixels is necessary.

Source: Automatic Trimap Generation for Image Matting

Image Source: Real-Time High-Resolution Background Matting

데이터셋

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

AIM-500

AIM-500 is the first natural image matting test set, contains 500 high-resolution real-world natural images from three types of images (salient opaque foregrounds, salient transparent/meticulous foregrounds, non-salient foregrounds), and multiple categories. The amount of each category is shown in the following table. | Portrait | Animal | Transparent | Plant | Furniture | Toy | Fruit | | :----:| :----: | :----: | :----: | :----: | :----: | :----: | | 100 | 200 | 34 | 75 | 45 | 36 | 10 |

Distinctions-646

Dinstinctions-646 are composed of 646 foreground images with manually annotated alpha mattes

Agriculture-Vision

A large-scale aerial farmland image dataset for semantic segmentation of agricultural patterns. Collects 94,986 high-quality aerial images from 3,432 farmlands across the US, where each image consists of RGB and Near-infrared (NIR) channels with resolution as high as 10 cm per pixel. Source: Agriculture-Vision: A Large Aerial Image Database for Agricultural Pattern Analysis

P3M-10k

P3M-10k contains 10421 high-resolution real-world face-blurred portrait images, along with their manually labeled alpha mattes. The Dataset is aimed to aid research efforts in the area of portrait image matting and related topics.

PPM-100

PPM is a portrait matting benchmark with the following characteristics: - Fine Annotation - All images are labeled and checked carefully. - Natural Background - All images use the original background without replacement. - Rich Diversity - The images cover full/half body and various postures. - High Resolution - The resolution of images is between 1080p and 4k. Dataset is created by authors of real-time matting model MODNet to measure performance for matting task.

AM-2K

AM-2k (Animal Matting 2,000 Dataset) consists of 2,000 high-resolution images collected and carefully selected from websites with open licenses. AM-2k contains 20 categories of animals including alpaca, antelope, bear, camel, cat, cattle, deer, dog, elephant, giraffe, horse, kangaroo, leopard, lion, monkey, rabbit, rhinoceros, sheep, tiger, zebra, each with 100 real-world images of various appearance and diverse backgrounds

withoutbg100 dataset

The withoutbg100 dataset consists of 100 image and alpha matte pairs. These pairs are chosen to represent a wide range of subjects and complexities, specifically crafted to enhance and test the capabilities of image background removal algorithms. The dataset includes images with complex elements such as fur and objects with varying transparency levels, providing a substantial challenge to even advanced matting techniques.

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