The Segment Anything Model (SAM) and CLIP are remarkable vision foundation models (VFMs). SAM, a prompt driven segmentation model, excels in segmentation tasks across diverse domains, while CLIP is renowned for its zero shot recognition capabilities. However, their unified potential has not yet been explored in medical image segmentation. To adapt SAM to medical imaging, existing methods primarily rely on tuning strategies that require extensive data or prior prompts tailored to the specific task, making it particularly challenging when only a limited number of data samples are available. This work presents an in depth exploration of integrating SAM and CLIP into a unified framework for medical image segmentation. Specifically, we propose a simple unified framework, SaLIP, for organ segmentation. Initially, SAM is used for part based segmentation within the image, followed by CLIP to retrieve the mask corresponding to the region of interest (ROI) from the pool of SAM generated masks. Finally, SAM is prompted by the retrieved ROI to segment a specific organ. Thus, SaLIP is training and fine tuning free and does not rely on domain expertise or labeled data for prompt engineering. Our method shows substantial enhancements in zero shot segmentation, showcasing notable improvements in DICE scores across diverse segmentation tasks like brain (63.46%), lung (50.11%), and fetal head (30.82%), when compared to un prompted SAM. Code and text prompts will be available online.
Existing domain adaptation (DA) methods often involve pre-training on the source domain and fine-tuning on the target domain. For multi-target domain adaptation, having a dedicated/separate fine-tuned network for each target domain, that retain all the pre-trained model parameters, is prohibitively expensive. To address this limitation, we propose Convolutional Low-Rank Adaptation (ConvLoRA). ConvLoRA freezes pre-trained model weights, adds trainable low-rank decomposition matrices to convolutional layers, and backpropagates the gradient through these matrices thus greatly reducing the number of trainable parameters. To further boost adaptation, we utilize Adaptive Batch Normalization (AdaBN) which computes target-specific running statistics and use it along with ConvLoRA. Our method has fewer trainable parameters and performs better or on-par with large independent fine-tuned networks (with less than 0.9% trainable parameters of the total base model) when tested on the segmentation of Calgary-Campinas dataset containing brain MRI images. Our approach is simple, yet effective and can be applied to any deep learning-based architecture which uses convolutional and batch normalization layers. Code is available at: https://github.com/aleemsidra/ConvLoRA.
Deep neural networks used in computer vision have been shown to exhibit many social biases such as gender bias. Vision Transformers (ViTs) have become increasingly popular in computer vision applications, outperforming Convolutional Neural Networks (CNNs) in many tasks such as image classification. However, given that research on mitigating bias in computer vision has primarily focused on CNNs, it is important to evaluate the effect of a different network architecture on the potential for bias amplification. In this paper we therefore introduce a novel metric to measure bias in architectures, Accuracy Difference. We examine bias amplification when models belonging to these two architectures are used as a part of large multimodal models, evaluating the different image encoders of Contrastive Language Image Pretraining which is an important model used in many generative models such as DALL-E and Stable Diffusion. Our experiments demonstrate that architecture can play a role in amplifying social biases due to the different techniques employed by the models for feature extraction and embedding as well as their different learning properties. This research found that ViTs amplified gender bias to a greater extent than CNNs
Chest X-rays have been widely used for COVID-19 screening; however, 3D computed tomography (CT) is a more effective modality. We present our findings on COVID-19 severity prediction from chest CT scans using the STOIC dataset. We developed an ensemble deep learning based model that incorporates multiple neural networks to improve predictions. To address data imbalance, we used slicing functions and data augmentation. We further improved performance using test time data augmentation. Our approach which employs a simple yet effective ensemble of deep learning-based models with strong test time augmentations, achieved results comparable to more complex methods and secured the fourth position in the STOIC2021 COVID-19 AI Challenge. Our code is available on online: at: https://github.com/aleemsidra/stoic2021- baseline-finalphase-main.
Generative multimodal models based on diffusion models have seen tremendous growth and advances in recent years. Models such as DALL-E and Stable Diffusion have become increasingly popular and successful at creating images from texts, often combining abstract ideas. However, like other deep learning models, they also reflect social biases they inherit from their training data, which is often crawled from the internet. Manually auditing models for biases can be very time and resource consuming and is further complicated by the unbounded and unconstrained nature of inputs these models can take. Research into bias measurement and quantification has generally focused on small single-stage models working on a single modality. Thus the emergence of multistage multimodal models requires a different approach. In this paper, we propose Multimodal Composite Association Score (MCAS) as a new method of measuring gender bias in multimodal generative models. Evaluating both DALL-E 2 and Stable Diffusion using this approach uncovered the presence of gendered associations of concepts embedded within the models. We propose MCAS as an accessible and scalable method of quantifying potential bias for models with different modalities and a range of potential biases.
Unsupervised anomaly detection (AD) is critical for a wide range of practical applications, from network security to health and medical tools. Due to the diversity of problems, no single algorithm has been found to be superior for all AD tasks. Choosing an algorithm, otherwise known as the Algorithm Selection Problem (ASP), has been extensively examined in supervised classification problems, through the use of meta-learning and AutoML, however, it has received little attention in unsupervised AD tasks. This research proposes a new meta-learning approach that identifies an appropriate unsupervised AD algorithm given a set of meta-features generated from the unlabelled input dataset. The performance of the proposed meta-learner is superior to the current state of the art solution. In addition, a mixed model statistical analysis has been conducted to examine the impact of the meta-learner components: the meta-model, meta-features, and the base set of AD algorithms, on the overall performance of the meta-learner. The analysis was conducted using more than 10,000 datasets, which is significantly larger than previous studies. Results indicate that a relatively small number of meta-features can be used to identify an appropriate AD algorithm, but the choice of a meta-model in the meta-learner has a considerable impact.
Over the years, the paradigm of medical image analysis has shifted from manual expertise to automated systems, often using deep learning (DL) systems. The performance of deep learning algorithms is highly dependent on data quality. Particularly for the medical domain, it is an important aspect as medical data is very sensitive to quality and poor quality can lead to misdiagnosis. To improve the diagnostic performance, research has been done both in complex DL architectures and in improving data quality using dataset dependent static hyperparameters. However, the performance is still constrained due to data quality and overfitting of hyperparameters to a specific dataset. To overcome these issues, this paper proposes random data augmentation based enhancement. The main objective is to develop a generalized, data-independent and computationally efficient enhancement approach to improve medical data quality for DL. The quality is enhanced by improving the brightness and contrast of images. In contrast to the existing methods, our method generates enhancement hyperparameters randomly within a defined range, which makes it robust and prevents overfitting to a specific dataset. To evaluate the generalization of the proposed method, we use four medical datasets and compare its performance with state-of-the-art methods for both classification and segmentation tasks. For grayscale imagery, experiments have been performed with: COVID-19 chest X-ray, KiTS19, and for RGB imagery with: LC25000 datasets. Experimental results demonstrate that with the proposed enhancement methodology, DL architectures outperform other existing methods. Our code is publicly available at: https://github.com/aleemsidra/Augmentation-Based-Generalized-Enhancement
Research into the classification of Image with Text (IWT) troll memes has recently become popular. Since the online community utilizes the refuge of memes to express themselves, there is an abundance of data in the form of memes. These memes have the potential to demean, harras, or bully targeted individuals. Moreover, the targeted individual could fall prey to opinion manipulation. To comprehend the use of memes in opinion manipulation, we define three specific domains (product, political or others) which we classify into troll or not-troll, with or without opinion manipulation. To enable this analysis, we enhanced an existing dataset by annotating the data with our defined classes, resulting in a dataset of 8,881 IWT or multimodal memes in the English language (TrollsWithOpinion dataset). We perform baseline experiments on the annotated dataset, and our result shows that existing state-of-the-art techniques could only reach a weighted-average F1-score of 0.37. This shows the need for a development of a specific technique to deal with multimodal troll memes.
Advanced Driver-Assistance Systems (ADAS) have been attracting attention from many researchers. Vision-based sensors are the closest way to emulate human driver visual behavior while driving. In this paper, we explore possible ways to use visual attention (saliency) for object detection and tracking. We investigate: 1) How a visual attention map such as a \emph{subjectness} attention or saliency map and an \emph{objectness} attention map can facilitate region proposal generation in a 2-stage object detector; 2) How a visual attention map can be used for tracking multiple objects. We propose a neural network that can simultaneously detect objects as and generate objectness and subjectness maps to save computational power. We further exploit the visual attention map during tracking using a sequential Monte Carlo probability hypothesis density (PHD) filter. The experiments are conducted on KITTI and DETRAC datasets. The use of visual attention and hierarchical features has shown a considerable improvement of $\approx$8\% in object detection which effectively increased tracking performance by $\approx$4\% on KITTI dataset.
This work tackles the Pixel Privacy task put forth by MediaEval 2019. Our goal is to manipulate images in a way that conceals them from automatic scene classifiers while preserving the original image quality. We use the fast gradient sign method, which normally has a corrupting influence on image appeal, and devise two methods to minimize the damage. The first approach uses a map of pixel locations that are either salient or flat, and directs perturbations away from them. The second approach subtracts the gradient of an aesthetics evaluation model from the gradient of the attack model to guide the perturbations towards a direction that preserves appeal. We make our code available at: https://git.io/JesXr.