Correlation of Object Detection Performance with Visual Saliency and Depth Estimation

As object detection techniques continue to evolve, understanding their relationships with complementary visual tasks becomes crucial for optimising model architectures and computational resources. This paper investigates the correlations between object detection accuracy and two fundamental visual tasks: depth prediction and visual saliency prediction. Through comprehensive experiments using state-of-the-art models (DeepGaze IIE, Depth Anything, DPT-Large, and Itti's model) on COCO and Pascal VOC datasets, we find that visual saliency shows consistently stronger correlations with object detection accuracy (mA$\rho$ up to 0.459 on Pascal VOC) compared to depth prediction (mA$\rho$ up to 0.283). Our analysis reveals significant variations in these correlations across object categories, with larger objects showing correlation values up to three times higher than smaller objects. These findings suggest incorporating visual saliency features into object detection architectures could be more beneficial than depth information, particularly for specific object categories. The observed category-specific variations also provide insights for targeted feature engineering and dataset design improvements, potentially leading to more efficient and accurate object detection systems.

Integrating Saliency Ranking and Reinforcement Learning for Enhanced Object Detection

With the ever-growing variety of object detection approaches, this study explores a series of experiments that combine reinforcement learning (RL)-based visual attention methods with saliency ranking techniques to investigate transparent and sustainable solutions. By integrating saliency ranking for initial bounding box prediction and subsequently applying RL techniques to refine these predictions through a finite set of actions over multiple time steps, this study aims to enhance RL object detection accuracy. Presented as a series of experiments, this research investigates the use of various image feature extraction methods and explores diverse Deep Q-Network (DQN) architectural variations for deep reinforcement learning-based localisation agent training. Additionally, we focus on optimising the detection pipeline at every step by prioritising lightweight and faster models, while also incorporating the capability to classify detected objects, a feature absent in previous RL approaches. We show that by evaluating the performance of these trained agents using the Pascal VOC 2007 dataset, faster and more optimised models were developed. Notably, the best mean Average Precision (mAP) achieved in this study was 51.4, surpassing benchmarks set by RL-based single object detectors in the literature.

An XAI Approach to Deep Learning Models in the Detection of Ductal Carcinoma in Situ

During the last decade or so, there has been an insurgence in the deep learning community to solve health-related issues, particularly breast cancer. Following the Camelyon-16 challenge in 2016, several researchers have dedicated their time to build Convolutional Neural Networks (CNNs) to help radiologists and other clinicians diagnose breast cancer. In particular, there has been an emphasis on Ductal Carcinoma in Situ (DCIS); the clinical term for early-stage breast cancer. Large companies have given their fair share of research into this subject, among these Google Deepmind who developed a model in 2020 that has proven to be better than radiologists themselves to diagnose breast cancer correctly. We found that among the issues which exist, there is a need for an explanatory system that goes through the hidden layers of a CNN to highlight those pixels that contributed to the classification of a mammogram. We then chose an open-source, reasonably successful project developed by Prof. Shen, using the CBIS-DDSM image database to run our experiments on. It was later improved using the Resnet-50 and VGG-16 patch-classifiers, analytically comparing the outcome of both. The results showed that the Resnet-50 one converged earlier in the experiments. Following the research by Montavon and Binder, we used the DeepTaylor Layer-wise Relevance Propagation (LRP) model to highlight those pixels and regions within a mammogram which contribute most to its classification. This is represented as a map of those pixels in the original image, which contribute to the diagnosis and the extent to which they contribute to the final classification. The most significant advantage of this algorithm is that it performs exceptionally well with the Resnet-50 patch classifier architecture.

Predicting Relative Depth between Objects from Semantic Features

Vision and language tasks such as Visual Relation Detection and Visual Question Answering benefit from semantic features that afford proper grounding of language. The 3D depth of objects depicted in 2D images is one such feature. However it is very difficult to obtain accurate depth information without learning the appropriate features, which are scene dependent. The state of the art in this area are complex Neural Network models trained on stereo image data to predict depth per pixel. Fortunately, in some tasks, its only the relative depth between objects that is required. In this paper the extent to which semantic features can predict course relative depth is investigated. The problem is casted as a classification one and geometrical features based on object bounding boxes, object labels and scene attributes are computed and used as inputs to pattern recognition models to predict relative depth. i.e behind, in-front and neutral. The results are compared to those obtained from averaging the output of the monodepth neural network model, which represents the state-of-the art. An overall increase of 14% in relative depth accuracy over relative depth computed from the monodepth model derived results is achieved.