Semantic Object Segmentation in Cultural Sites using Real and Synthetic Data

Deep learning has become the main way of the object detection task for autonomous vehicles. Meanwhile, this method typically requires vast amounts of training data to reach their full potential. However, collecting the data from real world and labeling manually is an expensive, time-consuming and error-prone process. Synthetic image has the potential to replace real image for training neural networks, because image creation and labeling annotations are free in this way. For the network trained by synthetic images, the reality gap between real and synthetic images is the main obstacle to use it in the real world. And most previous works are only devoted to generate synthetic images with a good performance on model training, but lack of analysis of the domain gap that affects the performance. This work designs a method of generating the real and synthetic images to analyze and reduce the reality gap between synthetic and real images for car detection. Firstly, this work put one single car with no-background in a random background image to generate real single car and synthetic single car images. In order to further reduce the domain gap in content level, this method keeps the car distribution in synthetic images is similar with the distribution of car in real world. For the purpose of reducing the domain gap in appearance level, the parameters of camera model are same as the camera parameters of image collecting cars and the image is rendered by using the PBRT(Physically Based Ray Tracing) when we generated the synthetic images. Secondly, by training the neural network of instance segmentation with different datasets, the across validation result proves that the reality gap between synthetic and real images is no more than the domain gap between real images. Thirdly, the training results of datasets with different samples diversity show that the diversity of the samples yields better generalization between different datasets for car detection which can effectively reduce the domain gap.

An unsupervised domain adaptation scheme for single-stage artwork recognition in cultural sites

Cyanobacterial Harmful Algal Blooms (CyanoHABs) in lakes and reservoirs have increased substantially in recent decades due to different environmental factors. Its early detection is a crucial issue to minimize health effects, particularly in potential drinking and recreational water bodies. The use of Autonomous Surface Vehicles (ASVs) equipped with machine vision systems (cameras) onboard, represents a useful alternative at this time. In this regard, we propose an image Semantic Segmentation approach based on Deep Learning with Convolutional Neural Networks (CNNs) for the early detection of CyanoHABs considering an ASV perspective. The use of these models is justified by the fact that with their convolutional architecture, it is possible to capture both, spectral and textural information considering the context of a pixel and its neighbors. To train these models it is necessary to have data, but the acquisition of real images is a difficult task, due to the capricious appearance of the algae on water surfaces sporadically and intermittently over time and after long periods of time, requiring even years and the permanent installation of the image capture system. This justifies the generation of synthetic data so that sufficiently trained models are required to detect CyanoHABs patches when they emerge on the water surface. The data generation for training and the use of the semantic segmentation models to capture contextual information determine the need for the proposal, as well as its novelty and contribution.Three datasets of images containing CyanoHABs patches are generated: (a) the first contains real patches of CyanoHABs as foreground and images of lakes and reservoirs as background, but with a limited number of examples; (b) the second, contains synthetic patches of CyanoHABs generated with state-of-the-art Style-based Generative Adversarial Network Adaptive Discriminator Augmentation (StyleGAN2-ADA) and Neural Style Transfer as foreground and images of lakes and reservoirs as background, and (c) the third set, is the combination of the previous two. Four model architectures for semantic segmentation (UNet++, FPN, PSPNet, and DeepLabV3+), with two encoders as backbone (ResNet50 and EfficientNet-b6), are evaluated from each dataset on real test images and different distributions. The results show the feasibility of the approach and that the UNet++ model with EfficientNet-b6, trained on the third dataset, achieves good generalization and performance for the real test images.

Infrared (IR) images are essential to improve the visibility of dark or camouflaged objects. Object recognition and segmentation based on a neural network using IR images provide more accuracy and insight than color visible images. But the bottleneck is the amount of relevant IR images for training. It is difficult to collect real-world IR images for special purposes, including space exploration, military and fire-fighting applications. To solve this problem, we created color visible and IR images using a Unity-based 3D game editor. These synthetically generated color visible and IR images were used to train cycle consistent adversarial networks (CycleGAN) to convert visible images to IR images. CycleGAN has the advantage that it does not require precisely matching visible and IR pairs for transformation training. In this study, we discovered that additional synthetic data can help improve CycleGAN performance. Neural network training using real data (N = 20) performed more accurate transformations than training using real (N = 10) and synthetic (N = 10) data combinations. The result indicates that the synthetic data cannot exceed the quality of the real data. Neural network training using real (N = 10) and synthetic (N = 100) data combinations showed almost the same performance as training using real data (N = 20). At least 10 times more synthetic data than real data is required to achieve the same performance. In summary, CycleGAN is used with synthetic data to improve the IR image conversion performance of visible images.

Large annotated datasets are required for training deep learning models, but in medical imaging data sharing is often complicated due to ethics, anonymization and data protection legislation. Generative AI models, such as generative adversarial networks (GANs) and diffusion models, can today produce very realistic synthetic images, and can potentially facilitate data sharing. However, in order to share synthetic medical images it must first be demonstrated that they can be used for training different networks with acceptable performance. Here, we therefore comprehensively evaluate four GANs (progressive GAN, StyleGAN 1–3) and a diffusion model for the task of brain tumor segmentation (using two segmentation networks, U-Net and a Swin transformer). Our results show that segmentation networks trained on synthetic images reach Dice scores that are 80%–90% of Dice scores when training with real images, but that memorization of the training images can be a problem for diffusion models if the original dataset is too small. Our conclusion is that sharing synthetic medical images is a viable option to sharing real images, but that further work is required. The trained generative models and the generated synthetic images are shared on AIDA data hub.

Image synthesis in magnetic resonance (MR) imaging has been an active area of research for more than ten years. MR image synthesis can be used to create images that were not acquired or replace images that are corrupted by artifacts, which can be of great benefit in automatic image analysis. Although synthetic images have been used with success in many applications, it is quite often true that they do not look like real images. In practice, an expert can usually distinguish synthetic images from real ones. Generative adversarial networks (GANs) have significantly improved the realism of synthetic images. However, we argue that further improvements can be made through the introduction of noise in the synthesis process, which better models the actual imaging process. Accordingly, we propose a novel approach that incorporates randomness into the model in order to better approximate the distribution of real MR images. Results show that the proposed method has comparable accuracy with the state-of-the-art approaches as measured by multiple similarity measurements while also being able to control the noise level in synthetic images. To further demonstrate the superiority of this model, we present results from a human observer study on synthetic images, which shows that our results capture the essential features of real MR images.

The ability to detect objects in cultural sites from the egocentric point of view of the user can enable interesting applications for both the visitors and the manager of the site. Unfortunately, current object detection algorithms have to be trained on large amounts of labeled data, the collection of which is costly and time-consuming. While synthetic data generated from the 3D model of the cultural site can be used to train object detection algorithms, a significant drop in performance is generally observed when such algorithms are deployed to work with real images. In this paper, we consider the problem of unsupervised domain adaptation for object detection in cultural sites. Specifically, we assume the availability of synthetic labeled images and real unlabeled images for training. To study the problem, we propose a dataset containing 75244 synthetic and 2190 real images with annotations for 16 different artworks. We hence investigate different domain adaptation techniques based on image-to-image translation and feature alignment. Our analysis points out that such techniques can be useful to address the domain adaptation issue, while there is still plenty of space for improvement on the proposed dataset. We release the dataset at our web page to encourage research on this challenging topic: https://iplab.dmi.unict.it/EGO-CH-OBJ-ADAPT/.

In the past, people usually employ the facial feature extraction and shallow learners such as decision trees, SVM, Naive Bayes, etc. to classify faces of different races. Deep learning usually takes lots of time to train. But with the advances in hardware and new algorithm proposed, the training time problem is gradually alleviated. The deep convolutional neural networks have good effect on images classification. In this paper, we use the deep convolutional neural networks to try to solve the problem of classification faces of different racial origin. Because the convolutional neural networks usually require a huge amount of data for training for good performance, such training set of real racial faces is not available to us. As a result of small set of real racial faces, this study proposes to incorporate synthetic facial images in our training set to sufficiently increase the size of the training set. To the best of our knowledge, this study is the first to propose to incorporate synthetic racial faces to train a deep convolutional neural network to classify real racial faces. We compare the performance of only employ synthetic facial images and mixtures of synthetic and real facial images in the training set. Our experiments show that training with only the real facial images (2,500 images) can achieve 91.25% accuracy in classifying faces of three different race origins. However, the classification when training with a mixture of 2,500 real facial images and 15,000 synthetic facial images can be further improved to 98.5% in accuracy.

Chest radiography is one of the most ubiquitous imaging modalities, playing an essential role in screening, diagnosis and disease management. However, chest radiography interpretation is a time-consuming and complex task, requiring the availability of experienced radiologists. As such, automated diagnosis systems for pathology detection have been proposed aiming to reduce the burden on radiologists and reduce variability in image interpretation. While promising results have been obtained, particularly since the advent of deep learning, there are significant limitations in the developed solutions, namely the lack of representative data for less frequent pathologies and the learning of biases from the training data, such as patient position, medical devices and other markers as proxies for certain pathologies. The lack of explainability is also a challenge for the adoption of these solutions in clinical practice.Generative adversarial networks could play a significant role as a solution for these challenges as they allow to artificially create new realistic images. This way, new synthetic chest radiography images could be used to increase the prevalence of less represented pathology classes and decrease model biases as well as improving the explainability of automatic decisions by generating samples that serve as examples or counter-examples to the image being analysed, ensuring patient privacy.In this study, a few-shot generative adversarial network is used to generate synthetic chest radiography images. A minimum Fréchet Inception Distance score of 17.83 was obtained, allowing to generate convincing synthetic images. Perceptual validation was then performed by asking multiple readers to classify a mixed set of synthetic and real images. An average accuracy of 83.5% was obtained but a strong dependency on reader experience level was observed. While synthetic images showed structural irregularities, the overall image sharpness was a major factor in the decision of readers. The synthetic images were then validated using a MobileNet abnormality classifier and it was shown that over 99% of images were classified correctly, indicating that the generated images were correctly interpreted by the classifier. Finally, the use of the synthetic images during training of a YOLOv5 pathology detector showed that the addition of the synthetic images led to an improvement of mean average precision of 0.05 across 14 pathologies.In conclusion, the usage of few-shot generative adversarial networks for chest radiography image generation was shown and tested in multiple scenarios, establishing a baseline for future experiments to increase the applicability of generative models in clinical scenarios of automatic CXR screening and diagnosis tools.

With the development of the information age, the awareness of archives informatization of governments, organizations and groups at all levels is constantly increasing. The key content extraction of scanned archive images has been widely concerned by researchers. Based on the deep learning method, we extract text and key elements (ID photo, seal, date of birth) by analyzing the archive images. Based on the real archive images data set and synthetic archive images data set, we use the PP-OCR [1] model to finish the text structured extraction and YOLOv5 [2] to train a target detection model of ID photos, seals, and dates of birth to achieve the extraction of key elements. Since the real data set cannot be made public, we use ordinary archive data for explanation.KeywordsArchive imagesSynthetic archive imagesPP-OCRYOLOv5Text structured extractionKey elements extraction

We present an Automatic License Plate Recognition system designed around Convolutional Neural Networks (CNNs) and trained over synthetic plate images. We first design CNNs suitable for plate and character detection, sharing a common architecture and training procedure. Then, we generate synthetic images that account for the varying illumination and pose conditions encountered with real plate images and we use exclusively such synthetic images to train our CNNs. Experiments with real vehicle images captured in natural light with commodity imaging systems show precision and recall in excess of 93% despite our networks are trained exclusively on synthetic images.

While neural symbolic methods demonstrate impressive performance in visual question answering on synthetic images, their performance suffers on real images. We identify that the long-tail distribution of visual concepts and unequal importance of reasoning steps in real data are the two key obstacles that limit the models' real-world potentials. To address these challenges, we propose a new paradigm, Calibrating Concepts and Operations (CCO), which enables neural symbolic models to capture underlying data characteristics and to reason with hierarchical importance. Specifically, we introduce an executor with learnable concept embedding magnitudes for handling distribution imbalance, and an operation calibrator for highlighting important operations and suppressing redundant ones.Our experiments show CCO substantially boosts the performance of neural symbolic methods on real images. By evaluating models on the real world dataset GQA, CCO helps the neural symbolic method NSCL outperforms its vanilla counterpart by 9.1% (from 47.0% to 56.1%); this result also largely reduces the performance gap between symbolic and non-symbolic methods. Additionally, we create a perturbed test set for better understanding and analyzing model performance on real images. Code is available at https://lizw14.github.io/project/ccosr.

This study discusses the scaling properties of the spatial distribution of the December 26, 2004, Sumatra aftershocks. We estimate the spatial correlation dimension D2 of the epicentral distribution of aftershocks recorded by a local network operated by Geological Survey of India. We estimate the value of D2 for five blocks in the source area by using generalized correlation integral approach. We assess its bias due to finite data points, scaling range, effects of location errors, and boundary effects theoretically and apply it to real data sets. The correlation dimension was computed both for real as well as synthetic data sets that include randomly generated point sets obtained using uniform distributions and mimicking the number of events and outlines of the effective areas filled with epicenters. On comparing the results from the real data and random point sets from simulations, we found the lower limit of bias in D2 estimates from limited data sets to be 0.26. Thus, the spatial variation in correlation dimensions among different blocks using local data sets cannot be directly compared unless the influence of bias in the real aftershock data set is taken into account. They cannot also be used to infer the geometry of the faults. We also discuss the results in order to add constraints on the use of synthetic data and of different approaches for uncertainty analysis on spatial variation of D2. A difference in D2 values, rather than their absolute values, among small blocks is of interest to local data sets, which are correlated with their seismic b values. Taking into account the possible errors and biases, the average D2 values vary from 1.05 to 1.57 in the Andaman–Nicobar region. The relative change in D2 values can be interpreted in terms of clustering and diffuse seismic activity associated with the low and high D2 values, respectively. Overall, a relatively high D2 and low b value is consistent with high-magnitude, diffuse activity in space in the source region of the 2004 Sumatra earthquake.

Egocentric visitor localization and artwork detection in cultural sites using synthetic data

A methodology for the simulation of infrared scenes is presented. The approach is based on combining synthetic and real data. The thermal signature of each object is first estimated using a thermal model. Environmental parameters which contribute to this signature are also considered. Then, specific techniques are used to extract textures from real images and to use these textures in the generation process. Experimental results showing purely synthetic images and synthetic textures images are presented.© (1993) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.