In this research, we propose a tiny image segmentation model, L^3U-net, that works on low-resource edge devices in real-time. We introduce a data folding technique that reduces inference latency by leveraging the parallel convolutional layer processing capability of the CNN accelerators. We also deploy the proposed model to such a device, MAX78000, and the results show that L^3U-net achieves more than 90% accuracy over two different segmentation datasets with 10 fps.
Progress in digital pathology is hindered by high-resolution images and the prohibitive cost of exhaustive localized annotations. The commonly used paradigm to categorize pathology images is patch-based processing, which often incorporates multiple instance learning (MIL) to aggregate local patch-level representations yielding image-level prediction. Nonetheless, diagnostically relevant regions may only take a small fraction of the whole tissue, and MIL-based aggregation operation assumes that all patch representations are independent and thus mislays the contextual information from adjacent cell and tissue microenvironments. Consequently, the computational resources dedicated to a specific region are independent of its information contribution. This paper proposes a transformer-based architecture specifically tailored for histopathological image classification, which combines fine-grained local attention with a coarse global attention mechanism to learn meaningful representations of high-resolution images at an efficient computational cost. More importantly, based on the observation above, we propose a novel mixing-based data-augmentation strategy, namely ScoreMix, by leveraging the distribution of the semantic regions of images during the training and carefully guiding the data mixing via sampling the locations of discriminative image content. Thorough experiments and ablation studies on three challenging representative cohorts of Haematoxylin & Eosin (H&E) tumour regions-of-interest (TRoIs) datasets have validated the superiority of our approach over existing state-of-the-art methods and effectiveness of our proposed components, e.g., data augmentation in improving classification performance. We also demonstrate our method's interpretability, robustness, and cross-domain generalization capability.
To bridge the gap between supervised semantic segmentation and real-world applications that acquires one model to recognize arbitrary new concepts, recent zero-shot segmentation attracts a lot of attention by exploring the relationships between unseen and seen object categories, yet requiring large amounts of densely-annotated data with diverse base classes. In this paper, we propose a new open-world semantic segmentation pipeline that makes the first attempt to learn to segment semantic objects of various open-world categories without any efforts on dense annotations, by purely exploiting the image-caption data that naturally exist on the Internet. Our method, Vision-language-driven Semantic Segmentation (ViL-Seg), employs an image and a text encoder to generate visual and text embeddings for the image-caption data, with two core components that endow its segmentation ability: First, the image encoder is jointly trained with a vision-based contrasting and a cross-modal contrasting, which encourage the visual embeddings to preserve both fine-grained semantics and high-level category information that are crucial for the segmentation task. Furthermore, an online clustering head is devised over the image encoder, which allows to dynamically segment the visual embeddings into distinct semantic groups such that they can be classified by comparing with various text embeddings to complete our segmentation pipeline. Experiments show that without using any data with dense annotations, our method can directly segment objects of arbitrary categories, outperforming zero-shot segmentation methods that require data labeling on three benchmark datasets.
In this work, we advance the neural head avatar technology to the megapixel resolution while focusing on the particularly challenging task of cross-driving synthesis, i.e., when the appearance of the driving image is substantially different from the animated source image. We propose a set of new neural architectures and training methods that can leverage both medium-resolution video data and high-resolution image data to achieve the desired levels of rendered image quality and generalization to novel views and motion. We demonstrate that suggested architectures and methods produce convincing high-resolution neural avatars, outperforming the competitors in the cross-driving scenario. Lastly, we show how a trained high-resolution neural avatar model can be distilled into a lightweight student model which runs in real-time and locks the identities of neural avatars to several dozens of pre-defined source images. Real-time operation and identity lock are essential for many practical applications head avatar systems.
Reducing the requirement for densely annotated masks in medical image segmentation is important due to cost constraints. In this paper, we consider the problem of inferring pixel-level predictions of brain lesions by only using image-level labels for training. By leveraging recent advances in generative diffusion probabilistic models (DPM), we synthesize counterfactuals of "How would a patient appear if X pathology was not present?". The difference image between the observed patient state and the healthy counterfactual can be used for inferring the location of pathology. We generate counterfactuals that correspond to the minimal change of the input such that it is transformed to healthy domain. This requires training with healthy and unhealthy data in DPMs. We improve on previous counterfactual DPMs by manipulating the generation process with implicit guidance along with attention conditioning instead of using classifiers. Code is available at https://github.com/vios-s/Diff-SCM.
Adversarial patch attacks are an emerging security threat for real world deep learning applications. We present Demasked Smoothing, the first approach (up to our knowledge) to certify the robustness of semantic segmentation models against this threat model. Previous work on certifiably defending against patch attacks has mostly focused on image classification task and often required changes in the model architecture and additional training which is undesirable and computationally expensive. In Demasked Smoothing, any segmentation model can be applied without particular training, fine-tuning, or restriction of the architecture. Using different masking strategies, Demasked Smoothing can be applied both for certified detection and certified recovery. In extensive experiments we show that Demasked Smoothing can on average certify 64% of the pixel predictions for a 1% patch in the detection task and 48% against a 0.5% patch for the recovery task on the ADE20K dataset.
In medical image analysis, the subtle visual characteristics of many diseases are challenging to discern, particularly due to the lack of paired data. For example, in mild Alzheimer's Disease (AD), brain tissue atrophy can be difficult to observe from pure imaging data, especially without paired AD and Cognitively Normal ( CN ) data for comparison. This work presents Disease Discovery GAN ( DiDiGAN), a weakly-supervised style-based framework for discovering and visualising subtle disease features. DiDiGAN learns a disease manifold of AD and CN visual characteristics, and the style codes sampled from this manifold are imposed onto an anatomical structural "blueprint" to synthesise paired AD and CN magnetic resonance images (MRIs). To suppress non-disease-related variations between the generated AD and CN pairs, DiDiGAN leverages a structural constraint with cycle consistency and anti-aliasing to enforce anatomical correspondence. When tested on the Alzheimer's Disease Neuroimaging Initiative ( ADNI) dataset, DiDiGAN showed key AD characteristics (reduced hippocampal volume, ventricular enlargement, and atrophy of cortical structures) through synthesising paired AD and CN scans. The qualitative results were backed up by automated brain volume analysis, where systematic pair-wise reductions in brain tissue structures were also measured
Large deformations of organs, caused by diverse shapes and nonlinear shape changes, pose a significant challenge for medical image registration. Traditional registration methods need to iteratively optimize an objective function via a specific deformation model along with meticulous parameter tuning, but which have limited capabilities in registering images with large deformations. While deep learning-based methods can learn the complex mapping from input images to their respective deformation field, it is regression-based and is prone to be stuck at local minima, particularly when large deformations are involved. To this end, we present Stochastic Planner-Actor-Critic (SPAC), a novel reinforcement learning-based framework that performs step-wise registration. The key notion is warping a moving image successively by each time step to finally align to a fixed image. Considering that it is challenging to handle high dimensional continuous action and state spaces in the conventional reinforcement learning (RL) framework, we introduce a new concept `Plan' to the standard Actor-Critic model, which is of low dimension and can facilitate the actor to generate a tractable high dimensional action. The entire framework is based on unsupervised training and operates in an end-to-end manner. We evaluate our method on several 2D and 3D medical image datasets, some of which contain large deformations. Our empirical results highlight that our work achieves consistent, significant gains and outperforms state-of-the-art methods.
We present an approach to efficiently and effectively adapt a masked image modeling (MIM) pre-trained vanilla Vision Transformer (ViT) for object detection, which is based on our two novel observations: (i) A MIM pre-trained vanilla ViT can work surprisingly well in the challenging object-level recognition scenario even with random sampled partial observations, e.g., only 25% ~ 50% of the input sequence. (ii) In order to construct multi-scale representations for object detection, a random initialized compact convolutional stem supplants the pre-trained large kernel patchify stem, and its intermediate features can naturally serve as the higher resolution inputs of a feature pyramid without upsampling. While the pre-trained ViT is only regarded as the third-stage of our detector's backbone instead of the whole feature extractor, resulting in a ConvNet-ViT hybrid architecture. The proposed detector, named MIMDet, enables a MIM pre-trained vanilla ViT to outperform hierarchical Swin Transformer by 2.3 box AP and 2.5 mask AP on COCO, and achieve even better results compared with other adapted vanilla ViT using a more modest fine-tuning recipe while converging 2.8x faster. Code and pre-trained models are available at \url{https://github.com/hustvl/MIMDet}.
Diffusion models are recent generative models that have shown great success in image generation with the state-of-the-art performance. However, only a few researches have been conducted for image manipulation with diffusion models. Here, we present a novel DiffusionCLIP which performs text-driven image manipulation with diffusion models using Contrastive Language-Image Pre-training (CLIP) loss. Our method has a performance comparable to that of the modern GAN-based image processing methods for in and out-of-domain image processing tasks, with the advantage of almost perfect inversion even without additional encoders or optimization. Furthermore, our method can be easily used for various novel applications, enabling image translation from an unseen domain to another unseen domain or stroke-conditioned image generation in an unseen domain, etc. Finally, we present a novel multiple attribute control with DiffusionCLIPby combining multiple fine-tuned diffusion models.