Camera Parameters Research Articles

Geometry-driven self-supervision for 3D human pose estimation

Although 3D human pose estimation has recently made strides, it is still difficult to precisely recreate a 3D human posture from a single image without the aid of 3D annotation for the following reasons. Firstly, the process of reconstruction inherently suffers from ambiguity, as multiple 3D poses can be projected onto the same 2D pose. Secondly, accurately measuring camera rotation without laborious camera calibration is a difficult task. While some approaches attempt to address these issues using traditional computer vision algorithms, they are not differentiable and cannot be optimized through training. This paper introduces two modules that explicitly leverage geometry to overcome these challenges, without requiring any 3D ground-truth or camera parameters. The first module, known as the relative depth estimation module, effectively mitigates depth ambiguity by narrowing down the possible depths for each joint to only two candidates. The second module, referred to as the differentiable pose alignment module, calculates camera rotation by aligning poses from different views. The use of these geometrically interpretable modules reduces the complexity of training and yields superior performance. By adopting our proposed method, we achieve state-of-the-art results on standard benchmark datasets, surpassing other self-supervised methods and even outperforming several fully-supervised approaches that heavily rely on 3D annotations.

Multi-view 3D reconstruction based on deep learning: A survey and comparison of methods

An important objective in computer vision is to analyze multiple images and subsequently reconstruct the shape and structure in 3D. Traditional multi-view 3D reconstruction techniques extract and match key features from images with known camera parameters. However, this approach is inefficient and fails to fully exploit the advantages of multi-view information. Advancements in deep learning have revolutionized multi-view 3D reconstruction by enabling end-to-end 3D shape inferencing without the need for sequential feature matching typically found in conventional algorithms. Recent rapid progress in this field necessitates a thorough review of current algorithms and provide insight into method of improving 3D reconstruction performance. This review classifies reconstruction algorithms according to their resultant model, including depth map, voxel, point cloud, mesh, and implicit surface. Additionally, this review encompasses the inclusion of frequently employed network training loss functions for network training, assessment metrics, and the incorporation of 3D datasets. Experimental results are also presented to assess the performance of different algorithms. Finally, the paper concludes with a summary, discussion of challenges, and potential future directions.

ERROR ANALYSIS OF VISUAL ODOMETRY FOR A SMALL SIZE UNMANNED AERIAL VEHICLE

Abstract. Usage of small size Unmanned Aerial Vehicle (UAV) with autonomous flight capability became widespread at both civil and military applications because of its low costs and ease of use. However, inertial or GPS based positioning and orientation sensors can be disabled at battlefields and the UAV losses its path in a short time. Visual odometry becomes a solution for the aforementioned hassle. Real-time detection of ground control points, matching the same control points at the stereo pair images and correcting the position and orientation data with least square adjustment are the challenges of visual odometry. In this study, error analysis of position and orientation of a UAV flying through mountainous region which cannot obtain any augmenting data for position and orientation is conducted by error propagation law. The flight of UAV is simulated with exact initial position and orientation data. Optic images of ground obtained by nadir directed camera are georeferenced using the digital elevation model of the region, interior and exterior camera parameters. The digital elevation model is used to correct the relief displacement. The stereo-pair image is obtained when the UAV flies 200 m and image coordinates of the common control points are detected by colinearity equations. Position and orientation of UAV is corrected by least square adjustment and the process is repeated for each obtained image acquired at 200 m interval. The analysis revealed that the precisions of horizontal and vertical positioning becomes 75 and 200 meter, respectively. The precision of roll angle becomes worse than 0.5 degree, while the precisions of pitch and yaw angle are estimate to be better that 0.1 degree.

A high accuracy method for the sintering condition recognition of rotary kiln

According to the different material sintering conditions, the sintering conditions of alumina rotary kiln can be divided into: super-heated, super-chilled, and normal. In this paper, based on Local Binary Pattern (LBP) and the primary-color method, a novel feature extraction method is proposed to obtain information about temperature in flame images without calibrating camera parameters. Through the analysis of the problem as well as the experimental phenomena, a new classification procedure is devised: in the first step, the super-chilled condition is first separated, in the second step, the normal and super-heat condition are classified. Different feature extraction methods are used in the two steps mentioned above. One is to extract the texture features of pseudo temperature images, and short-time energy is used to describe the dynamic features. The other is to extract the texture features of grey-images by our improved LBPP,Rriu2, and characterize the dynamics with sample entropy and variance. Finally, experimental results show that the our feature extraction method can effectively reduce intra-class variation, increase inter-class variation and receive a high classification accuracy.

A calibration and prediction method of the camera imaging parameters in variable temperature environment

In the on-orbit application of photogrammetry, constraints on thermal control resources or aging of thermal control equipment will trigger increased fluctuations in the camera operating environment temperature. Consequently, the parameters of the camera imaging model vary, such that the measurement accuracy is reduced. Accordingly, the accurate prediction of the camera imaging parameters (e.g., Camera Interior Parameters (CIP) and Lens Distortion Parameters (LDP)) at variable temperatures takes on critical significance.First, a novel method, termed as Virtual Multi-station Camera Calibration (VMCC), is proposed to calibrate the CIP and LDP with each frame at variable temperatures. Second, the Variational Mode Decomposition method combined with the Northern Goshawk Optimizer (NGO-VMD) is employed to decompose the camera imaging parameter signals obtained from VMCC, yielding the n Intrinsic Mode Functions (IMFs) contained in the respective parameter signal. Third, the Generalized Regression Neural Network combined with the Dung Beetle Optimizer (DBO-GRNN) is employed, such that the smoothness parameter q of the GRNN is iteratively optimized based on the camera imaging parameter prediction error. Lastly, the CIP and the LDP are accurately predicted at different temperatures in accordance with the temperature data from multiple parts of the camera.Based on this study, a reliable dataset of camera imaging parameter responses at variable temperatures can be generated for the on-orbit photogrammetry system. Furthermore, the prediction accuracy and immense application potential of neural network modeling methods are demonstrated in camera parameter prediction.

Spatial quadric calibration method for multi-line laser based on diffractive optical element

The traditional multi-line laser calibration method relies primarily on calibrating the spatial plane equations of multiple laser lines and reconstructing the multi-line laser in three dimensions through the plane equation and camera parameters. In traditional methods, the multi-line laser is projected as a straight line by default, but in reality, the laser line projected by the multi-line laser is not a plane due to the Diffractive Optical Element (DOE). Instead, it is a surface with a certain curvature in space. During the production process, the multi-line laser experiences significant distortion because of the DOE projecting at a large angle. Consequently, if traditional methods are used, accurate calibration of the multi-line laser cannot be achieved. To address this issue, this paper introduces a new calibration algorithm for multi-line laser based on a spatial quadric equation. By calibrating the spatial quadric equation of the multi-line laser, the spatial characteristics of the laser can be accurately captured. The algorithm involves projecting a crossed 7-line laser onto a calibration plate and changing the position of the calibration plate while capturing images. This allows for obtaining the spatial three-dimensional coordinates of each laser line in the camera coordinate system. Next, spatial quadric equations are fitted for each laser line based on the three-dimensional coordinates. Finally, 3D reconstruction is performed using the quadratic surface parameters of the multi-line laser and camera parameters. The results from 3D reconstruction demonstrate that, compared to traditional laser calibration methods based on planes, the proposed algorithm achieves higher calibration accuracy. It also improves the number of successful binocular multi-line laser matches and overall accuracy.

INFLUENCE ANALYSIS OF THREE DIFFERENT CAMERA CALIBRATION METHODS ON A CAMERA-PROJECTOR MEASURING SYSTEM

The challenges arising from mass production with high quality level require advanced error measurement systems and optical metrology systems. Nowadays, it is more frequent to find this type of controls at the manufacturing industry into the quality control procedures as long as the need for a faster capture speed and precision has increased. However, these techniques can be improved. The accuracy of 3D measurement systems formed by a camera-projector pair depends directly on the calibration procedure used. The projector is usually modelled as the inverse projection of a pin-hole camera, which presents the option to calibrate the system by two different approaches: as a whole system or separately. The most common approach is the use of a camera previously calibrated, followed by the calibration of the projector. Studies show that the uncertainty of the camera parameters from its calibration propagates to the projector parameters and several authors conclude that the three most widespread are: Tsai, Zhang and Direct Linear Calibration. The objective of this study is to have a clear comprehension of the relationships between the camera and projector parameters and of how uncertainty is propagated to the measurement system error. Therefore, the three calibration methods previously mentioned and some of them possible combinations are analyzed.

A correction method for radial distortion and nonlinear response of infrared cameras.

The key feature of non-contact temperature measurement provided by infrared (IR) cameras underpins their versatility. However, the accuracy of temperature measurements with IR cameras depends on imaging quality due to their non-contact nature, such as the lens, body temperature, and measurement environment. This paper addresses the correction of radial distortion and nonlinear response issues in IR cameras. To address radial distortion, we have designed a passive checkerboard calibration board specifically for infrared cameras. This board is used to calibrate the IR camera and derive the necessary camera parameters. Subsequently, these parameters are applied during the actual measurement process to rectify radial distortion effectively. Building on the radial distortion correction method mentioned above, we propose a multi-point segmented calibration approach that considers different temperature ranges and imaging regions. This method alleviates the issue of reduced temperature measurement accuracy due to variations in camera responses by computing gain and offset coefficient matrices for each temperature range. Experimental results demonstrate the effectiveness of the calibration board in correcting radial distortion in IR cameras, with a mean reprojection error of less than 0.16 pixels. Regarding the nonlinear response problem, the introduced method significantly reduces the relative error in temperature measurement. In the verification phase, spanning from 100 to 500 °C, the average relative error in temperature measurement decreases by 0.49% from 1.61% before and after correction, which highlights a substantial improvement in temperature measurement accuracy. This work gives a useful reference to improve the imaging quality and temperature measurement accuracy using infrared cameras.

A multi‐degree‐of‐freedom monitoring method for slope displacement based on stereo vision

AbstractThree‐dimensional displacement monitoring over long distances has been a long‐standing concern in the structural health monitoring industry. In this study, a multi‐degree‐of‐freedom slope displacement monitoring method is developed by fusing computer vision and the 3D point triangulation method. Attributed to this method, the problems of outdoor binocular camera calibration, multi‐target mismatching, and outdoor illumination effects were solved. First, a two‐stage camera calibration method is proposed to accurately calibrate intrinsic and extrinsic camera parameters under a large field of view and long working distance conditions. Second, the adaptive spatial‐frequency method is proposed to calculate the coding and pixel coordinates of the monitored target. In this step, to solve the problem of mismatching monitored points in different camera frames, the Augmented Reality University of Cordoba code is introduced to provide a unique identity code for each monitored point. To mitigate the impact of illumination and other factors on pixel coordinate calculation, an adaptive pixel coordinate calculation method that combines information from the spatial and frequency domains is proposed., Third, based on the intrinsic and extrinsic parameters of the stereo camera and the pixel coordinates of the monitored points, the 3D coordinates of the monitored points are obtained through triangulation. Finally, the accuracy experiments and stability experiments are conducted. According to the results of the experiments, the measurement distance is positively correlated with the measurement error. And the baseline length is negatively correlated with the measurement error in the z‐direction. Ultimately, we suggest that the ratio of baseline length to measurement distance should be greater than 40%. When the recommended value is satisfied, the measurement error is less than 1 mm when the measurement distance is less than 40 m. When the measurement distance is equal to 90 m, the measurement error is less than 5 mm. Meanwhile, stability experiments of the algorithm were carried out, and in a period of outdoor validation experiments, the fluctuations were only sub‐millimeter, demonstrating good anti‐interference performance. Moreover, the method proposed in this study successfully monitored a landslide disaster in Guangxi, which demonstrated its outstanding practical application capabilities.

Iterative Camera Calibration Method Based on Concentric Circle Grids

A concentric circle target is commonly used in the vision measurement system for its detection accuracy and robustness. To enhance the camera calibration accuracy, this paper proposes an improved calibration method that utilizes concentric circle grids as the calibration target. The method involves accurately locating the imaged center and optimizing camera parameters. The imaged concentric circle center obtained by cross-ratio invariance is not affected by perspective projection, which ensures the location accuracy of the feature point. Subsequently, the impact of lens distortion on camera calibration is comprehensively investigated. The sub-pixel coordinates of imaged centers are taken into the iterative calibration method, and camera parameters are updated. Through simulations and real experiments, the proposed method effectively reduces the residual error and improves the accuracy of camera parameters.

Progressive Learning of 3D Reconstruction Network From 2D GAN Data.

This paper presents a method to reconstruct high-quality textured 3D models from single images. Current methods rely on datasets with expensive annotations; multi-view images and their camera parameters. Our method relies on GAN generated multi-view image datasets which have a negligible annotation cost. However, they are not strictly multi-view consistent and sometimes GANs output distorted images. This results in degraded reconstruction qualities. In this work, to overcome these limitations of generated datasets, we have two main contributions which lead us to achieve state-of-the-art results on challenging objects: 1) A robust multi-stage learning scheme that gradually relies more on the models own predictions when calculating losses and 2) A novel adversarial learning pipeline with online pseudo-ground truth generations to achieve fine details. Our work provides a bridge from 2D supervisions of GAN models to 3D reconstruction models and removes the expensive annotation efforts. We show significant improvements over previous methods whether they were trained on GAN generated multi-view images or on real images with expensive annotations.

Evaluating SLAM Performance with Synthesized Datasets from Unreal-Based Emulators

The rapid advancement in visual-inertial simultaneous localization and mapping (SLAM) has opened numerous applications in computer vision. However, the scarcity of high quality, publicly accessible datasets hampers the evaluation of SLAM performance in varied and tailored environments. In this study, I employed the AirSim simulator and the Unreal Engine 4 to generate a trajectory resembling that of the TUM VI Room 1 ground truth dataset within the ArchViz indoor environment representing a well-lit, furnished room. I further modified the environment and trajectory through various expansions, addition of features, and data smoothing to ensure a more stable sequence of input frames into the SLAM architecture. I then examined the efficiency of visual ORB-SLAM3 by inputting images of resolution 256×144 and 512×288 at 30 frames per second (FPS), while also adjusting the feature threshold - the maximum number of feature points that ORB-SLAM3 tracks per frame. This investigation of the camera parameters within AirSim and ORB-SLAM3 has led to the essential finding that the resolution of the input images must coincide with the dimensions of the film. The subsequent runs under these variables reveal that higher resolution images lead to considerably better tracking, with an optimal feature threshold ranging between 3000~12000 feature points per frame. Moreover, ORB- SLAM3 demonstrated significantly enhanced robustness within dynamic environments containing moving objects when using higher resolution inputs, with a decreased error of close to 0cm compared to 23.19cm for lower resolutions (averaged over three runs). Finally, I conducted qualitative testing using real-life indoor environments recorded with an iPhone Xr camera, which produces results that highlight the challenges faced by ORB-SLAM3 due to factors such as glare and motion blur.

Proximity-Based Optical Camera Communication with Multiple Transmitters Using Deep Learning.

In recent years, optical camera communication (OCC) has garnered attention as a research focus. OCC uses optical light to transmit data by scattering the light in various directions. Although this can be advantageous with multiple transmitter scenarios, there are situations in which only a single transmitter is permitted to communicate. Therefore, this method is proposed to fulfill the latter requirement using 2D object size to calculate the proximity of the objects through an AI object detection model. This approach enables prioritization among transmitters based on the transmitter proximity to the receiver for communication, facilitating alternating communication with multiple transmitters. The image processing employed when receiving the signals from transmitters enables communication to be performed without the need to modify the camera parameters. During the implementation, the distance between the transmitter and receiver varied between 1.0 and 5.0 m, and the system demonstrated a maximum data rate of 3.945 kbps with a minimum BER of 4.2×10-3. Additionally, the system achieved high accuracy from the refined YOLOv8 detection algorithm, reaching 0.98 mAP at a 0.50 IoU.

Synthetic Training Data in AI-Driven Quality Inspection: The Significance of Camera, Lighting, and Noise Parameters.

Industrial-quality inspections, particularly those leveraging AI, require significant amounts of training data. In fields like injection molding, producing a multitude of defective parts for such data poses environmental and financial challenges. Synthetic training data emerge as a potential solution to address these concerns. Although the creation of realistic synthetic 2D images from 3D models of injection-molded parts involves numerous rendering parameters, the current literature on the generation and application of synthetic data in industrial-quality inspection scarcely addresses the impact of these parameters on AI efficacy. In this study, we delve into some of these key parameters, such as camera position, lighting, and computational noise, to gauge their effect on AI performance. By utilizing Blender software, we procedurally introduced the "flash" defect on a 3D model sourced from a CAD file of an injection-molded part. Subsequently, with Blender's Cycles rendering engine, we produced datasets for each parameter variation. These datasets were then used to train a pre-trained EfficientNet-V2 for the binary classification of the "flash" defect. Our results indicate that while noise is less critical, using a range of noise levels in training can benefit model adaptability and efficiency. Variability in camera positioning and lighting conditions was found to be more significant, enhancing model performance even when real-world conditions mirror the controlled synthetic environment. These findings suggest that incorporating diverse lighting and camera dynamics is beneficial for AI applications, regardless of the consistency in real-world operational settings.

An Image Processing Approach for Real-Time Safety Assessment of Autonomous Drone Delivery

The aim of producing self-driving drones has driven many researchers to automate various drone driving functions, such as take-off, navigation, and landing. However, despite the emergence of delivery as one of the most important uses of autonomous drones, there is still no automatic way to verify the safety of the delivery stage. One of the primary steps in the delivery operation is to ensure that the dropping zone is a safe area on arrival and during the dropping process. This paper proposes an image-processing-based classification approach for the delivery drone dropping process at a predefined destination. It employs live streaming via a single onboard camera and Global Positioning System (GPS) information. A two-stage processing procedure is proposed based on image segmentation and classification. Relevant parameters such as camera parameters, light parameters, dropping zone dimensions, and drone height from the ground are taken into account in the classification. The experimental results indicate that the proposed approach provides a fast method with reliable accuracy based on low-order calculations.

CoBEV: Elevating Roadside 3D Object Detection with Depth and Height Complementarity.

Roadside camera-driven 3D object detection is a crucial task in intelligent transportation systems, which extends the perception range beyond the limitations of vision-centric vehicles and enhances road safety. While previous studies have limitations in using only depth or height information, we find both depth and height matter and they are in fact complementary. The depth feature encompasses precise geometric cues, whereas the height feature is primarily focused on distinguishing between various categories of height intervals, essentially providing semantic context. This insight motivates the development of Complementary-BEV (CoBEV), a novel end-to-end monocular 3D object detection framework that integrates depth and height to construct robust BEV representations. In essence, CoBEV estimates each pixel's depth and height distribution and lifts the camera features into 3D space for lateral fusion using the newly proposed two-stage complementary feature selection (CFS) module. A BEV feature distillation framework is also seamlessly integrated to further enhance the detection accuracy from the prior knowledge of the fusion-modal CoBEV teacher. We conduct extensive experiments on the public 3D detection benchmarks of roadside camera-based DAIR-V2X-I and Rope3D, as well as the private Supremind-Road dataset, demonstrating that CoBEV not only achieves the accuracy of the new state-of-the-art, but also significantly advances the robustness of previous methods in challenging long-distance scenarios and noisy camera disturbance, and enhances generalization by a large margin in heterologous settings with drastic changes in scene and camera parameters. For the first time, the vehicle AP score of a camera model reaches 80% on DAIR-V2X-I in terms of easy mode. The source code will be made publicly available at CoBEV.

Estimation of planar angles from non-orthogonal imaging.

Photogrammetry-based methods are traditionally used to estimate the geometrical parameters using optical images. These methods employ specific equipment, computationally sophisticated and expensive algorithms, and utilize projective geometry to reconstruct real-life scenes up to a scale. In this work, we used a computationally less-expensive method for sparse reconstruction to estimate the planar angles using two-view geometry and linear algorithms from non-orthogonal images acquired by a smartphone camera. First, intrinsic camera parameters were determined. Next, scale-invariant feature transform was used to identify the correspondence points from each pair of images. Epipolar constraint was applied on all these points to determine the essential matrix using the eight-point algorithm. Thereafter, extrinsic camera parameters were estimated from the essential matrix and combined with the intrinsic matrix to get the camera projection matrix. Finally, linear triangulation was used to get the sparse point cloud representing the scene. Planar angles were estimated by backprojecting the chosen image points and applying simple vector algebra on the obtained 3D points. The method was successful in estimating the planar angles in less than 10s on non-curved edges with an average error of 3% by using only ten images. Given the simplicity of methods used, this technique can be integrated into a smartphone for on-site measurements as well as large deformations.

Portable paper-based probe for on-site ratiometric fluorescence determination of total flavonol glycosides in plant extract using smartphone imaging.

A smartphone-assisted, paper-based ratio fluorescence probe is presented for the rapid, low-cost and on-site quantification of total flavonol glycosides in Ginkgo biloba extracts (GBE). The Al3+/Eu-MOF/paper-based probe utilizes lanthanide metal-organic framework (Ln-MOF) nanoparticles immobilized on Whatman filter paper along with Al3+ for detecting flavonols, which are the hydrolyzed products of flavonol glycosides. The color change of the paper-based fluorescence image from red to orange depends on the concentration of the target analyte in the sample solution. The smartphone equipped with a red, green, blue (RGB) color detector measured the fluorescence signal intensity on the paper substrate after adding flavonol. The analytical variables affecting the performance of the probe, including the addition sequence of the aluminum nitrate solution, its concentration, that of the Ln-MOF solution, the drying time of the paper probe, the reaction time and the sensitivity parameters of the mobile phone camera (ISO), were optimized. Under optimal conditions, the Al3+/Eu-MOF/paper-based probe has good linear response in the concentration range 7 ~ 80µg mL- 1 and a lower detection limit of 2.07µg mL- 1. The results obtained with the paper-based ratio fluorescence probe and smartphone combination were validated by comparing them with high-performance liquid chromatography (HPLC) measurements. This study provides a potential strategy for fabricating Al3+/Eu-MOF/paper-based probe used for total flavonol glycosides determination.

Glutathione amino acid precursors protect skin from UVB-induced damage and improve skin tone.

UV radiation exposure causes skin irritation, erythema, darkening and barrier disruption by inducing oxidative stress and inflammation. Glutathione, a master antioxidant, plays an important role in the antioxidant defence network of the skin. This study aimed to assess the invitro protective effects of the glutathione amino acid precursors blend (GAP) on transcriptomic and phenotypic endpoints against UVB-induced challenges. Normal human epidermal melanocytes (NHEMs) were exposed to GAP, ascorbic acid (AA) and its derivatives. Viability was assessed using the CCK8 method. Melakutis®, a pigmented living skin equivalent (pLSE) model, underwent repeated 50 mJ/cm2 UVB irradiation with or without GAP treatment. Images of the model were captured with consistent camera parameters, and the model's light intensity was measured using a spectrophotometer. Melanin content was determined by measuring absorbance at 405 nm. Confirmation of melanin deposition and distribution was achieved through Fontana-Masson staining. Transcriptomic analysis was conducted using RNA sequencing (RNA-Seq), and a machine learning approach was employed for transcriptomic aging clock analysis. In NHEMs, all tested compounds exhibited over 85% viability compared to the vehicle control, indicating no heightened risk of cytotoxicity. Notably, GAP demonstrated greater efficacy in inhibiting melanin production than AA derivatives at equivalent concentrations. In pLSE models, GAP notably enhanced model lightness, and reduced melanin content and deposition following the UVB challenge, whereas AA showed minimal impact. GAP effectively counteracted UVB-induced alterations in gene expression linked to pigmentation, inflammation and aging. Moreover, recurrent UVB exposure substantially elevated the biological age of pLSE models, a phenomenon mitigated by GAP treatment. In NHEMs, GAP exhibited enhanced effectiveness in inhibiting melanin production at identical tested doses in comparison to AA derivatives. Noteworthy protective effects of GAP against UVB irradiation were observed in the pLSE models, as evidenced by skin pigmentation measurements and transcriptomic changes.

SyDog-Video: A Synthetic Dog Video Dataset for Temporal Pose Estimation

We aim to estimate the pose of dogs from videos using a temporal deep learning model as this can result in more accurate pose predictions when temporary occlusions or substantial movements occur. Generally, deep learning models require a lot of data to perform well. To our knowledge, public pose datasets containing videos of dogs are non existent. To solve this problem, and avoid manually labelling videos as it can take a lot of time, we generated a synthetic dataset containing 500 videos of dogs performing different actions using Unity3D. Diversity is achieved by randomising parameters such as lighting, backgrounds, camera parameters and the dog’s appearance and pose. We evaluate the quality of our synthetic dataset by assessing the model’s capacity to generalise to real data. Usually, networks trained on synthetic data perform poorly when evaluated on real data, this is due to the domain gap. As there was still a domain gap after improving the quality of the synthetic dataset and inserting diversity, we bridged the domain gap by applying 2 different methods: fine-tuning and using a mixed dataset to train the network. Additionally, we compare the model pre-trained on synthetic data with models pre-trained on a real-world animal pose datasets. We demonstrate that using the synthetic dataset is beneficial for training models with (small) real-world datasets. Furthermore, we show that pre-training the model with the synthetic dataset is the go to choice rather than pre-training on real-world datasets for solving the pose estimation task from videos of dogs.