Image Processing

With the development and implementation of convolutional neural networks in pattern recognition, there are large number of parameters needs to calculate and storage, which makes the algorithm hard to run in common computer.


This dataset provides RGB and Depth images acquired by Kinect v2 of 10 cerebral palsy patients. For each subject (0001, 0002, ecc) there are 12 folders: 

- 5 folders containing 5 left full gait cycles (L_01, L_02, ecc)

- 5 folders containing 5 right full gait cycles (R_01, R_02. ecc)

- 1 folder containing one static lateral view (left side) of the subject while standing upright (L_s)

- 1 folder containing one static lateral view (right side) of the subject while standing upright  (R_s)

In each folder (dynamic and static) there are two subfolders:


The dataset was generated through the execution of a Python script designed to collect a comprehensive set of data samples from six different sensors for each specific gesture. Upon launching the script, users are prompted to initiate gesture 0, Once ready, users can commence recording, with the program automatically capturing 1000 samples for that particular gesture. Subsequently, the program prompts users to perform gesture 1, and this process repeats until data for all gestures is collected.


This dataset collects samples of different kinds of defective and normal chenille yarn images for the same batch of chenille yarn made of polyester material, aiming to facilitate the task of recognizing and classifying chenille yarn defects in computer vision and machine learning algorithms. This dataset consists of a total of 2500 images of 5 major chenille yarn defects and 2500 normal chenille yarn images, totaling 5000 images. It is captured by an industrial camera in the state of chenille yarn movement.


Lettuce Farm SLAM Dataset (LFSD) is a VSLAM dataset based on RGB and depth images captured by VegeBot robot in a lettuce farm. The dataset consists of RGB and depth images, IMU, and RTK-GPS sensor data. Detection and tracking of lettuce plants on images are annotated with the standard Multiple Object Tracking (MOT) format. It aims to accelerate the development of algorithms for localization and mapping in the agricultural field, and crop detection and tracking.


The demand for artificial intelligence (AI) in healthcare is rapidly increasing. However, significant challenges arise from data scarcity and privacy concerns, particularly in medical imaging. While existing generative models have achieved success in image synthesis and image-to-image translation tasks, there remains a gap in the generation of 3D semantic medical images. To address this gap, we introduce Med-DDPM, a diffusion model specifically designed for semantic 3D medical image synthesis, effectively tackling data scarcity and privacy issues.


This dataset consists of a test result dataset with 10 sample images and a test result dataset with a artificial image. Backpropagation neural networks (BPNNs) can be used to restore images; however, the error surface of the BPNN algorithm contains several extrema, making it easy to slip into a locally optimal solution. A genetic algorithm (GA) with a strong global searchability can optimize the initial weight and threshold of BPNNs. However, traditional GAs are prone to local convergence and stagnation; hence, we propose a hybrid GA.


This dataset contains 37 estrogen receptor immunohistochemistry (ER-IHC) whole slide images (WSIs) obtained from Universiti Malaya Medical Centre (UMMC), Malaysia. The WSI is scanned using 3DHistech Pannoramic DESK at 20x magnification with an approximate dimension of 80,000 pixels width and 200,000 pixels height per WSI.


Realistic benchmark datasets are crucial for providing a consistent measurement baseline for comparing different BlindSR methods and testing their generalization ability in real-world scenarios. However, the evaluation of BlindSR methods is still limited by the lack of a common dataset that conforms to realistic degradation scenarios. We construct a benchmark dataset that follows real scenarios, which reflects the real-world BlindSR problem more accurately than existing synthetic datasets.


A novel method is proposed in the paper to obtain the high-precision estimation of the angular velocity and the star-spot’s centroid at the same time. First, the Radon transform on a single frame image is adopted to roughly estimate the initial value of angular velocity and the star-spot’s centroid based on the dynamic imaging and kinematic model. Then, theloss function of each star spot is constructed by combining two consecutive frames.