Incremental Learning

The CIFAR-100 dataset (Canadian Institute for Advanced Research, 100 classes) is a subset of the Tiny Images dataset and consists of 60000 32x32 color images. The 100 classes in the CIFAR-100 are grouped into 20 superclasses. There are 600 images per class. Each image comes with a "fine" label (the class to which it belongs) and a "coarse" label (the superclass to which it belongs). There are 500 training images and 100 testing images per class. The criteria for deciding whether an image belongs to a class were as follows: * The class name should be high on the list of likely answers to the question “What is in this picture?” * The image should be photo-realistic. Labelers were instructed to reject line drawings. * The image should contain only one prominent instance of the object to which the class refers. * The object may be partially occluded or seen from an unusual viewpoint as long as its identity is still clear to the labeler. Source: https://www.cs.toronto.edu/~kriz/cifar.html Image Source: https://www.cs.toronto.edu/~kriz/cifar.html

CORe50 is a dataset designed for assessing Continual Learning techniques in an Object Recognition context. Source: CORe50: a New Dataset and Benchmark for Continuous Object Recognition

Permuted MNIST is an MNIST variant that consists of 70,000 images of handwritten digits from 0 to 9, where 60,000 images are used for training, and 10,000 images for test. The difference of this dataset from the original MNIST is that each of the ten tasks is the multi-class classification of a different random permutation of the input pixels. Source: Lifelong Learning with Dynamically Expandable Networks Image Source: https://arxiv.org/pdf/1810.12488.pdf

Collected from top 10 most popular clothing/wearable brandname logos captured in rich visual context. Source: Deep Learning Logo Detection with Data Expansion by Synthesising Context

Click to add a brief description of the dataset (Markdown and LaTeX enabled). Provide: * a high-level explanation of the dataset characteristics * explain motivations and summary of its content * potential use cases of the dataset

Concept-1K contains 1023 novel concepts from six domains, including economy, culture, science and technology, environment, education, and health and medical. It has 16653 training-test QA pairs corresponding to 16653 knowledge points from 1023 concepts. It is proposed for evaluating the forgetting in large language models and the effectiveness of incremental learning algorithms.

HOWS-CL-25 (Household Objects Within Simulation dataset for Continual Learning) is a synthetic dataset especially designed for object classification on mobile robots operating in a changing environment (like a household), where it is important to learn new, never seen objects on the fly. This dataset can also be used for other learning use-cases, like instance segmentation or depth estimation. Or where household objects or continual learning are of interest. Our dataset contains 150,795 unique synthetic images using 25 different household categories with 925 3D models in total. For each of those categories, we generated about 6000 RGB images. In addition, we also provide a corresponding depth, segmentation, and normal image. The dataset was created with BlenderProc [Denninger et al. (2019)], a procedural pipeline to generate images for deep learning. This tool created a virtual room with randomly textured floors, walls, and a light source with randomly chosen light intensity and color. After that, a 3D model is placed in the resulting room. This object gets customized by randomly assigning materials, including different textures, to achieve a diverse dataset. Moreover, each object might be deformed with a random displacement texture. We use 774 3D models from the ShapeNet dataset [A. X. Chang et al. (2015)] and the other models from various internet sites. Please note that we had to manually fix and filter most of the models with Blender before using them in the pipeline! For continual learning (CL), we provide two different loading schemes: - Five sequences with five categories each - Twelve sequences with three categories in the first and two in the other sequences. In addition to the RGB, depth, segmentation, and normal images, we also provide the calculated features of the RGB images (by ResNet50) as used in our RECALL paper. In those two loading schemes, ten percent of the images are used for validation, where we ensure that an object instance is either in the training or the validation set, not in both. This avoids learning to recognize certain instances by heart. We recommend using those loading schemes to compare your approach with others. For further information and code examples, please have a look at our website: https://github.com/DLR-RM/RECALL.

Simulates unanticipated user needs in the deployment stage. Source: Incremental Learning from Scratch for Task-Oriented Dialogue Systems