Deep segmentation models often face the failure risks when the testing image presents unseen distributions. Improving model robustness against these risks is crucial for the large-scale clinical application of deep models. In this study, inspired by human learning cycle, we propose a novel online reflective learning framework (RefSeg) to improve segmentation robustness. Based on the reflection-on-action conception, our RefSeg firstly drives the deep model to take action to obtain semantic segmentation. Then, RefSeg triggers the model to reflect itself. Because making deep models realize their segmentation failures during testing is challenging, RefSeg synthesizes a realistic proxy image from the semantic mask to help deep models build intuitive and effective reflections. This proxy translates and emphasizes the segmentation flaws. By maximizing the structural similarity between the raw input and the proxy, the reflection-on-action loop is closed with segmentation robustness improved. RefSeg runs in the testing phase and is general for segmentation models. Extensive validation on three medical image segmentation tasks with a public cardiac MR dataset and two in-house large ultrasound datasets show that our RefSeg remarkably improves model robustness and reports state-of-the-art performance over strong competitors.
As sharing images in an instant message is a crucial factor, there has been active research on learning a image-text multi-modal dialogue model. However, training a well-generalized multi-modal dialogue model is challenging because existing multi-modal dialogue datasets contain a small number of data, limited topics, and a restricted variety of images per dialogue. In this paper, we present a multi-modal dialogue dataset creation pipeline that involves matching large-scale images to dialogues based on CLIP similarity. Using this automatic pipeline, we propose a large-scale multi-modal dialogue dataset, DialogCC, which covers diverse real-world topics and various images per dialogue. With extensive experiments, we demonstrate that training a multi-modal dialogue model with our dataset can improve generalization performance. Additionally, existing models trained with our dataset achieve state-of-the-art performance on image and text retrieval tasks. The source code and the dataset will be released after publication.
The neuron reconstruction from raw Optical Microscopy (OM) image stacks is the basis of neuroscience. Manual annotation and semi-automatic neuron tracing algorithms are time-consuming and inefficient. Existing deep learning neuron reconstruction methods, although demonstrating exemplary performance, greatly demand complex rule-based components. Therefore, a crucial challenge is designing an end-to-end neuron reconstruction method that makes the overall framework simpler and model training easier. We propose a Neuron Reconstruction Transformer (NRTR) that, discarding the complex rule-based components, views neuron reconstruction as a direct set-prediction problem. To the best of our knowledge, NRTR is the first image-to-set deep learning model for end-to-end neuron reconstruction. In experiments using the BigNeuron and VISoR-40 datasets, NRTR achieves excellent neuron reconstruction results for comprehensive benchmarks and outperforms competitive baselines. Results of extensive experiments indicate that NRTR is effective at showing that neuron reconstruction is viewed as a set-prediction problem, which makes end-to-end model training available.
We explore the generation of visualisations of audio latent spaces using an audio-to-image generation pipeline. We believe this can help with the interpretability of audio latent spaces. We demonstrate a variety of results on the NSynth dataset. A web demo is available.
Unsupervised anomaly detection in medical imaging aims to detect and localize arbitrary anomalies without requiring annotated anomalous data during training. Often, this is achieved by learning a data distribution of normal samples and detecting anomalies as regions in the image which deviate from this distribution. Most current state-of-the-art methods use latent variable generative models operating directly on the images. However, generative models have been shown to mostly capture low-level features, s.a. pixel-intensities, instead of rich semantic features, which also applies to their representations. We circumvent this problem by proposing CRADL whose core idea is to model the distribution of normal samples directly in the low-dimensional representation space of an encoder trained with a contrastive pretext-task. By utilizing the representations of contrastive learning, we aim to fix the over-fixation on low-level features and learn more semantic-rich representations. Our experiments on anomaly detection and localization tasks using three distinct evaluation datasets show that 1) contrastive representations are superior to representations of generative latent variable models and 2) the CRADL framework shows competitive or superior performance to state-of-the-art.
An important goal of self-supervised learning is to enable model pre-training to benefit from almost unlimited data. However, one method that has recently become popular, namely masked image modeling (MIM), is suspected to be unable to benefit from larger data. In this work, we break this misconception through extensive experiments, with data scales ranging from 10\% of ImageNet-1K to full ImageNet-22K, model sizes ranging from 49 million to 1 billion, and training lengths ranging from 125K iterations to 500K iterations. Our study reveals that: (i) Masked image modeling is also demanding on larger data. We observed that very large models got over-fitted with relatively small data; (ii) The length of training matters. Large models trained with masked image modeling can benefit from more data with longer training; (iii) The validation loss in pre-training is a good indicator to measure how well the model performs for fine-tuning on multiple tasks. This observation allows us to pre-evaluate pre-trained models in advance without having to make costly trial-and-error assessments of downstream tasks. We hope that our findings will advance the understanding of masked image modeling in terms of scaling ability.
Developing a deep learning method for medical segmentation tasks heavily relies on a large amount of labeled data. However, the annotations require professional knowledge and are limited in number. Recently, semi-supervised learning has demonstrated great potential in medical segmentation tasks. Most existing methods related to cardiac magnetic resonance images only focus on regular images with similar domains and high image quality. A semi-supervised domain generalization method was developed in [2], which enhances the quality of pseudo labels on varied datasets. In this paper, we follow the strategy in [2] and present a domain generalization method for semi-supervised medical segmentation. Our main goal is to improve the quality of pseudo labels under extreme MRI Analysis with various domains. We perform Fourier transformation on input images to learn low-level statistics and cross-domain information. Then we feed the augmented images as input to the double cross pseudo supervision networks to calculate the variance among pseudo labels. We evaluate our method on the CMRxMotion dataset [1]. With only partially labeled data and without domain labels, our approach consistently generates accurate segmentation results of cardiac magnetic resonance images with different respiratory motions. Code will be available after the conference.
Fundamental differences between natural and medical images have recently favored the use of self-supervised learning (SSL) over ImageNet transfer learning for medical image applications. Differences between image types are primarily due to the imaging modality and medical images utilize a wide range of physics based techniques while natural images are captured using only visible light. While many have demonstrated that SSL on medical images has resulted in better downstream task performance, our work suggests that more performance can be gained. The scientific principles which are used to acquire medical images are not often considered when constructing learning problems. For this reason, we propose incorporating quantitative imaging principles during generative SSL to improve image quality and quantitative biological accuracy. We show that this training schema results in better starting states for downstream supervised training on limited data. Our model also generates images that validate on clinical quantitative analysis software.
Realistic long-horizon tasks like image-goal navigation involve exploratory and exploitative phases. Assigned with an image of the goal, an embodied agent must explore to discover the goal, i.e., search efficiently using learned priors. Once the goal is discovered, the agent must accurately calibrate the last-mile of navigation to the goal. As with any robust system, switches between exploratory goal discovery and exploitative last-mile navigation enable better recovery from errors. Following these intuitive guide rails, we propose SLING to improve the performance of existing image-goal navigation systems. Entirely complementing prior methods, we focus on last-mile navigation and leverage the underlying geometric structure of the problem with neural descriptors. With simple but effective switches, we can easily connect SLING with heuristic, reinforcement learning, and neural modular policies. On a standardized image-goal navigation benchmark (Hahn et al. 2021), we improve performance across policies, scenes, and episode complexity, raising the state-of-the-art from 45% to 55% success rate. Beyond photorealistic simulation, we conduct real-robot experiments in three physical scenes and find these improvements to transfer well to real environments.
Our aim is to build autonomous agents that can solve tasks in environments like Minecraft. To do so, we used an imitation learning-based approach. We formulate our control problem as a search problem over a dataset of experts' demonstrations, where the agent copies actions from a similar demonstration trajectory of image-action pairs. We perform a proximity search over the BASALT MineRL-dataset in the latent representation of a Video PreTraining model. The agent copies the actions from the expert trajectory as long as the distance between the state representations of the agent and the selected expert trajectory from the dataset do not diverge. Then the proximity search is repeated. Our approach can effectively recover meaningful demonstration trajectories and show human-like behavior of an agent in the Minecraft environment.