A pothole video dataset for semantic segmentation.

This paper introduces a video dataset for semantic segmentation of road potholes. This dataset contains 619 high-resolution videos captured in January 2023, covering locations in eight villages within the Hulu Sungai Tengah regency of South Kalimantan, Indonesia. The dataset is divided into three main folders, namely train, val, and test. The train, val, and test folders contain 372 videos for training, 124 videos for validation, and 123 videos for testing, respectively. Each of these main folders has two subfolders, ``RGB'' for the video in the RGB format and ``mask'' for the ground truth segmentation. These videos are precisely two seconds long, containing 48 frames each, and all are in MP4 format. The dataset offers remarkable flexibility, accommodating various research needs, from full-video segmentation to frame extraction. It enables researchers to create ground truth annotations and change the combination of videos in the folders according to their needs. This resource is an asset for researchers, engineers, policymakers, and anyone interested in advancing algorithms for pothole detection and analysis. This dataset allows for benchmarking semantic segmentation algorithms, conducting comparative studies on pothole detection methods, and exploring innovative approaches, offering valuable contributions to the computer vision community.

Efficient and Robust Method to Detect the Location of Macular Center Based on Optimal Temporal Determination.

The location of the macular central is very important for the examination of macular edema when using an automated screening system. The erratic character of the macular light intensity and the absence of a clear border make this anatomical structure difficult to detect. This paper presents a new method for detecting the macular center based on its geometrical location in the temporal direction of the optic disc. Also, a new method of determining the temporal direction using the vascular features visible on the optic disc is proposed. After detecting the optic disc, the temporal direction is determined by considering blood vessel positions. The macular center is detected using thresholding and simple morphology operations with optimum macular region of interest (ROI) direction. The results show that the proposed method has a low computation time of 0.34 s/image with 100% accuracy for the DRIVE dataset, while that of DiaretDB1 was 0.57 s/image with 98.87% accuracy.

CNN-Based Classification for Highly Similar Vehicle Model Using Multi-Task Learning.

Vehicle make and model classification is crucial to the operation of an intelligent transportation system (ITS). Fine-grained vehicle information such as make and model can help officers uncover cases of traffic violations when license plate information cannot be obtained. Various techniques have been developed to perform vehicle make and model classification. However, it is very hard to identify the make and model of vehicles with highly similar visual appearances. The classifier contains a lot of potential for mistakes because the vehicles look very similar but have different models and manufacturers. To solve this problem, a fine-grained classifier based on convolutional neural networks with a multi-task learning approach is proposed in this paper. The proposed method takes a vehicle image as input and extracts features using the VGG-16 architecture. The extracted features will then be sent to two different branches, with one branch being used to classify the vehicle model and the other to classify the vehicle make. The performance of the proposed method was evaluated using the InaV-Dash dataset, which contains an Indonesian vehicle model with a highly similar visual appearance. The experimental results show that the proposed method achieves 98.73% accuracy for vehicle make and 97.69% accuracy for vehicle model. Our study also demonstrates that the proposed method is able to improve the performance of the baseline method on highly similar vehicle classification problems.

Erythrocyte (red blood cell) dataset in thalassemia case.

Red blood cell (RBC) dataset was obtained from four thalassemia peripheral blood smears and a healthy peripheral blood smear. The dataset contains 7108 images of individual red blood cells for nine cell types. The first process is image acquisition, which is the process of retrieving microscopic image data from peripheral blood smears through a Olympus CX21 microscope using an Optilab advance plus camera. Laboratory assistants helped obtain ideal erythrocyte images. We provide peripheral blood smear from four thalassemia patients in the ThalassemiaPBS dataset. After image acquisition, the image is resized from 4100 × 3075 pixels to 800 × 600 pixels to reduce the computing load in the next stage. We extracted the green color component (green channel) of the RGB image and used it in the next process. We chose the green channel because it is not affected by variations in color and brightness. Furthermore, the segmentation stage is carried out to obtain an object in the form of a single red blood cell. After that, the object can be classified according to the type of red blood cell. This dataset can become an opportunity for international researchers to develop the classification method for red blood cells.

Region-based annotation data of fire images for intelligent surveillance system.

This paper presents fire segmentation annotation data on 12 commonly used and publicly available "VisiFire Dataset" videos from http://signal.ee.bilkent.edu.tr/VisiFire/. This annotations dataset was obtained by per-frame, manual hand annotation over the fire region with 2684 total annotated frames. Since this annotation provides per-frame segmentation data, it offers a new and unique fire motion feature to the existing video, unlike other fire segmentation data that are collected from different still images. The annotations dataset also provides ground truth for segmentation task on videos. With segmentation task, it offers better insight on how well a machine learning model understood, not only detecting whether a fire is present, but also its exact location by calculating metrics such as Intersection over Union (IoU) with this annotations data. This annotations data is a tremendously useful addition to train, develop, and create a much better smart surveillance system for early detection in high-risk fire hotspots area.

Registered Relief Depth (RRD) borobudur dataset for single-frame depth prediction on one-side artifacts.

Single-frame depth prediction is an efficient 3D reconstruction method for one-side artifacts. However, for this purpose, ground truth images, where the pixels are associated with the actual depth, are needed. The small number of publicly accessible datasets is an issue with the restoration of cultural heritage objects. In addition, relief data with irregular characteristics due to nature and human treatment, such as decolorization caused by moss and chemical reaction is still not available. We therefore created a dataset of Borobudur temple reliefs registered with their depth for data availability to solve these problems. This data collection consists of 4608 × 3456 (4K) resolution and profound RGB frames and we call this dataset the Registered Relief Depth (RRD) Borobudur Dataset. The RGB images have been taken using an Olympus EM10 II Camera with a 14 mm f/3.5 lens and the depth images were obtained directly using an ASUS XTION scanner, acquired on the temple's reliefs at 15000-25000 lux day time. The registration process of RGB data and depth information was manually performed via control points and was directly supervised by the archaeologist. Apart of enriching the data availability, this dataset can become an opportunity for International researchers to understand more about Indonesian Cultural Heritages.

Automated Detection of Retinal Nerve Fiber Layer by Texture-Based Analysis for Glaucoma Evaluation.

OBJECTIVES: The retinal nerve fiber layer (RNFL) is a site of glaucomatous optic neuropathy whose early changes need to be detected because glaucoma is one of the most common causes of blindness. This paper proposes an automated RNFL detection method based on the texture feature by forming a co-occurrence matrix and a backpropagation neural network as the classifier. METHODS: We propose two texture features, namely, correlation and autocorrelation based on a co-occurrence matrix. Those features are selected by using a correlation feature selection method. Then the backpropagation neural network is applied as the classifier to implement RNFL detection in a retinal fundus image. RESULTS: We used 40 retinal fundus images as testing data and 160 sub-images (80 showing a normal RNFL and 80 showing RNFL loss) as training data to evaluate the performance of our proposed method. Overall, this work achieved an accuracy of 94.52%. CONCLUSIONS: Our results demonstrated that the proposed method achieved a high accuracy, which indicates good performance.