In the Pascal VOC2012 dataset, semantic segmentation labels are provided in an image format that perfectly matches the spatial dimensions (height and width) of the original input images. Each pixel in a label image represents a semantic class by its RGB color; pixels of the same color belong to the same category. By convention in this dataset, black pixels designate the background, white pixels indicate borders between objects, and various other distinct colors correspond to the predefined target classes (such as aeroplane, bicycle, or bird).

Pascal VOC2012 Semantic Segmentation Labels

In semantic segmentation tasks, the input image and its corresponding ground-truth label maintain a strict one-to-one spatial correspondence at the pixel level. Because of this precise alignment, simply rescaling images to fit a model's required input shape is problematic; it requires inversely rescaling the predicted pixel classes back to the original dimensions during inference, which introduces inaccuracies along the boundaries of different semantic regions. To avoid these artifacts and preserve exact pixel correspondence, the input images and their labels are typically subjected to random fixed-shape cropping rather than rescaling.

Fixed-Shape Cropping in Semantic Segmentation

The Pascal VOC2012 dataset serves as one of the most critical and widely adopted benchmarks for semantic segmentation tasks in computer vision. It features a comprehensive collection of images paired with detailed pixel-level annotations. Structurally, the dataset is organized into specific directories: ImageSets/Segmentation contains the text files defining the training and validation data splits, JPEGImages stores the original input photographs, and SegmentationClass houses the corresponding segmented label images.

Claude

In computer vision there is many established datasets used for training and benchmarking different algorithms. Unlike other machine learning areas, computer vision datasets are much more standardized because it takes significantly more effort to create them.

Computer Vision Datasets

Semantic segmentation is a computer vision task focused on recognizing and understanding the contents of an image at a granular, pixel level. It operates by dividing the entire input image into distinct regions, assigning every individual pixel to a specific, predefined semantic class (such as a car, pedestrian, or background).

Semantic Segmentation

Dive into Deep Learning

The most popular 68 facial landmarks detection in real-world datasets of facial images captured in the wild. Reannotated versions of the LFPW, AFW, HELEN, and XM2VTS datasets along with an additional 135 images.

300-W (300 Faces in the Wild)

Provides an accessible image database that's organized relative to the WordNet hierarchy. WordNet has over 100,000 synsets (synonym rings) which is a group of words that are semantically equivalent. ImageNet provides 1,000 images to illustrate each synset. All of this totals to over 10 million cleanly sorted images with the WordNet concepts (more than just labeling it an image a word).

ImageNet

Offers visual relationship annotations indicating pairs of objects in particular relations, object properties, and human actions. All object bounding boxes are manually drawn by professional annotators and every scene has diverse and often complex scenes with multiple objects. What makes Open Images different from other object classifiers is it has relations in its labels such as "woman playing guitar", "table is wooden", etc instead of single word classifications of objects.

Open Images

Contains a diverse set of stereo video sequences recorded in street scenes from 50 different cities. Includes annotations for 5,000 frames at high-quality pixel level and 20,000 weakly annotated frames. It is mainly used for training and benchmarking neural networks pertaining to semantic urban scene understanding.

Cityscapes

Consists of labeled subsets of tiny images (32x32 pixels) labeled in 10 different image classes. The dataset is divided into five training batches and one test batch. All the classes are mutually exclusive meaning there is no overlap in their meaning.

CIFAR-10

Consists of 1593 handwritten digits from around 80 people. Each person wrote all the digits between 0 and 9 twice and were instructed to write it normally then quickly.

Semeion Handwritten Digit Data set

Consists of Zalando's article images with 60k examples for training and 10k for testing. It was created because the MNIST dataset has become to simple for modern CNNs and it's one of the most popular datasets to benchmarks algorithms on. Fashion-MNIST seeks to provide a new benchmark.

Fashion-MNIST

Large-scale dataset of videos that include a diverse range of human focused actions. The dataset consists of around 650,000 videos covering 700 human action classes with a least 600 videos for each. The clips last around 10 seconds and are labelled with a single class.

Kinetics-700

Large scale object detection, segmentation, and captioning dataset. Abbreviation is Common Objects in COntext.

MS COCO

25K images with over 40K people with annotated body joints. All images are extracted from a YouTube video and provided with preceding and following un-annotated frames.

MPII Human Pose Dataset

The banana detection dataset is a collection of data used specifically for object detection tasks. It is publicly accessible and consists of image files along with corresponding CSV label files that provide the necessary annotations. This dataset can be downloaded directly from the internet and is often used to demonstrate and practice deep learning object detection algorithms.

Learn Before

Related

Learn After