Abstract: Researchers have extensively studied the field of vision and language, discovering that both visual and textual content is crucial for understanding scenes effectively. Particularly, comprehending text in videos holds great significance, requiring both scene text understanding and temporal reasoning. This paper focuses on exploring two recently introduced datasets, NewsVideoQA and M4-ViteVQA, which aim to address video question answering based on textual content. The NewsVideoQA dataset contains question-answer pairs related to the text in news videos, while M4- ViteVQA comprises question-answer pairs from diverse categories like vlogging, traveling, and shopping. We provide an analysis of the formulation of these datasets on various levels, exploring the degree of visual understanding and multi-frame comprehension required for answering the questions. Additionally, the study includes experimentation with BERT-QA, a text-only model, which demonstrates comparable performance to the original methods on both datasets, indicating the shortcomings in the formulation of these datasets. Furthermore, we also look into the domain adaptation aspect by examining the effectiveness of training on M4-ViteVQA and evaluating on NewsVideoQA and vice-versa, thereby shedding light on the challenges and potential benefits of out-of-domain training.

Abstract: In recent years, deep learning models have considerably advanced the performance of segmentation tasks on Brain Magnetic Resonance Imaging (MRI). However, these models show a considerable performance drop when they are evaluated on unseen data from a different distribution. Since annotation is often a hard and costly task requiring expert supervision, it is necessary to develop ways in which existing models can be adapted to the unseen domains without any additional labelled information. In this work, we explore one such technique which extends the CycleGAN [2] architecture to generate label-preserving data in the target domain. The synthetic target domain data is used to train the nn-UNet [3] framework for the task of multi-label segmentation. The experiments are conducted and evaluated on the dataset [1] provided in the ‘Cross-Modality Domain Adaptation for Medical Image Segmentation’ challenge [23] for segmentation of vestibular schwannoma (VS) tumour and cochlea on contrast enhanced (ceT1) and high resolution (hrT2) MRI scans. In the proposed approach, our model obtains dice scores (DSC) 0.73 and 0.49 for tumour and cochlea respectively on the validation set of the dataset. This indicates the applicability of the proposed technique to real-world problems where data may be obtained by different acquisition protocols as in [1] where hrT2 images are more reliable, safer, and lower-cost alternative to ceT1.

Abstract: Several recent works on self-supervised learning are trained by mapping different augmentations of the same image to the same feature representation. The data augmentations used are of crucial importance to the quality of learned feature representations. In this paper, we analyze how the color jitter traditionally used in data augmentation negatively impacts the quality of the color features in learned feature representations. To address this problem, we propose a more realistic, physics-based color data augmentation – which we call Planckian Jitter – that creates realistic variations in chromaticity and produces a model robust to illumination changes that can be commonly observed in real life, while maintaining the ability to discriminate image content based on color information. Experiments confirm that such a representation is complementary to the representations learned with the currently-used color jitter augmentation and that a simple concatenation leads to significant performance gains on a wide range of downstream datasets. In addition, we present a color sensitivity analysis that documents the impact of different training methods on model neurons and shows that the performance of the learned features is robust with respect to illuminant variations.

Abstract: An intraluminal coronary stent is a metal scaold deployed in a stenotic artery during Percutaneous Coronary Intervention (PCI).
Intravascular Ultrasound (IVUS) is a catheter-based imaging technique generally used for assessing the correct placement of the stent. All the approaches proposed so far for the stent analysis only focused on the struts detection, while this paper proposes a novel approach to detect the boundaries and the position of the stent along the pullback.
The pipeline of the method requires the identication of the stable frames
of the sequence and the reliable detection of stent struts. Using this data,
a measure of likelihood for a frame to contain a stent is computed. Then,
a robust binary representation of the presence of the stent in the pullback
is obtained applying an iterative and multi-scale approximation of the signal to symbols using the SAX algorithm. Results obtained comparing the automatic results versus the manual annotation of two observers on 80 IVUS in-vivo sequences shows that the method approaches the inter-observer variability scores.

Abstract: The assessment of arterial luminal area, performed by IVUS analysis, is a clinical index used to evaluate the degree of coronary artery disease. In this paper we propose a novel approach to automatically segment the vessel lumen, which combines model-based temporal information extracted from successive frames of the sequence, with spatial classification using the Growcut algorithm. The performance of the method is evaluated by an in vivo experiment on 300 IVUS frames. The automatic and manual segmentation performances in general vessel and stent frames are comparable. The average segmentation error in vessel, stent and bifurcation frames are 0.17±0.08 mm, 0.18±0.07 mm and 0.31±0.12 mm respectively.

Abstract: Workshop STACOM, dins del MICCAI
X-ray angiography is the most common imaging modality employed in the diagnosis of coronary diseases prior to or during a catheter-based intervention. The analysis of the patient X-Ray sequence can provide useful information about the degree of arterial stenosis, the myocardial perfusion and other clinical parameters. If the sequence has been acquired to evaluate the perfusion grade, the opacity due to the diaphragm could potentially hinder any kind of visual inspection and make more difficult a computer aided measurements. In this paper we propose an accurate and robust method to automatically identify the diaphragm border in each frame. Quantitative evaluation on a set of 11 sequences shows that the proposed algorithm outperforms previous methods.

Abstract: A signed distance function (SDF) as the 3D shape description is one of the most effective approaches to represent 3D geometry for rendering and reconstruction. Our work is inspired by the state-of-the-art method DeepSDF [17] that learns and analyzes the 3D shape as the iso-surface of its shell and this method has shown promising results especially in the 3D shape reconstruction and compression domain. In this paper, we consider the degeneration problem of reconstruction coming from the capacity decrease of the DeepSDF model, which approximates the SDF with a neural network and a single latent code. We propose Local Geometry Code Learning (LGCL), a model that improves the original DeepSDF results by learning from a local shape geometry of the full 3D shape. We add an extra graph neural network to split the single transmittable latent code into a set of local latent codes distributed on the 3D shape. Mentioned latent codes are used to approximate the SDF in their local regions, which will alleviate the complexity of the approximation compared to the original DeepSDF. Furthermore, we introduce a new geometric loss function to facilitate the training of these local latent codes. Note that other local shape adjusting methods use the 3D voxel representation, which in turn is a problem highly difficult to solve or even is insolvable. In contrast, our architecture is based on graph processing implicitly and performs the learning regression process directly in the latent code space, thus make the proposed architecture more flexible and also simple for realization. Our experiments on 3D shape reconstruction demonstrate that our LGCL method can keep more details with a significantly smaller size of the SDF decoder and outperforms considerably the original DeepSDF method under the most important quantitative metrics.

Abstract: Domain adaptation (DA) aims to transfer the knowledge learned from a source domain to an unlabeled target domain. Some recent works tackle source-free domain adaptation (SFDA) where only a source pre-trained model is available for adaptation to the target domain. However, those methods do not consider keeping source performance which is of high practical value in real world applications. In this paper, we propose a new domain adaptation paradigm called Generalized Source-free Domain Adaptation (G-SFDA), where the learned model needs to perform well on both the target and source domains, with only access to current unlabeled target data during adaptation. First, we propose local structure clustering (LSC), aiming to cluster the target features with its semantically similar neighbors, which successfully adapts the model to the target domain in the absence of source data. Second, we propose sparse domain attention (SDA), it produces a binary domain specific attention to activate different feature channels for different domains, meanwhile the domain attention will be utilized to regularize the gradient during adaptation to keep source information. In the experiments, for target performance our method is on par with or better than existing DA and SFDA methods, specifically it achieves state-of-the-art performance (85.4%) on VisDA, and our method works well for all domains after adapting to single or multiple target domains.