Donggeun Yoo

Cedric Caruzzo, Donggeun Yoo, Tae Soo Kim

Retrieval-augmented generation evaluation checks whether model claims are factually grounded in retrieved documents. It does not check whether retrieved evidence is attributed to the correct entity. A clinical RAG response can pass every automated check (zero hallucinations, near-perfect faithfulness, real citations) while presenting drug Y's clinical evidence as evidence about queried drug X. We term this deceptive grounding (DG): a failure invisible to faithfulness, hallucination, and citation checks because every claim is sourced from a real document, about the wrong entity. Using a controlled factorial benchmark across 13 models, we find DG rates spanning 8-87% at peak adversarial conditions. Medical and biomedical fine-tuned models reach up to 86.7%; domain specialization amplifies the failure rather than mitigating it. A controlled ablation identifies the mechanism: removing entity-specific clinical evidence from retrieved documents eliminates entity-attribution failure entirely, shifting all failures to confabulation. The two failure modes respond to the same trigger, taking different paths. Production measurement across 740 drug-disease pairs finds 7.8% overall DG in a deployed RAG system, rising to 13.6% for recently approved drugs. Entity-attribution verification (checking that cited evidence applies to the queried entity) detects DG at 97.0% precision and 98.7% DG recall (IPW-adjusted human gold standard); no existing framework implements it.

Minchul Kim, Jongchan Park, Seil Na et al.

Finding diseases from an X-ray image is an important yet highly challenging task. Current methods for solving this task exploit various characteristics of the chest X-ray image, but one of the most important characteristics is still missing: the necessity of comparison between related regions in an image. In this paper, we present Attend-and-Compare Module (ACM) for capturing the difference between an object of interest and its corresponding context. We show that explicit difference modeling can be very helpful in tasks that require direct comparison between locations from afar. This module can be plugged into existing deep learning models. For evaluation, we apply our module to three chest X-ray recognition tasks and COCO object detection & segmentation tasks and observe consistent improvements across tasks. The code is available at https://github.com/mk-minchul/attend-and-compare.

Hyeonseob Nam, HyunJae Lee, Jongchan Park et al.

Convolutional Neural Networks (CNNs) often fail to maintain their performance when they confront new test domains, which is known as the problem of domain shift. Recent studies suggest that one of the main causes of this problem is CNNs' strong inductive bias towards image styles (i.e. textures) which are sensitive to domain changes, rather than contents (i.e. shapes). Inspired by this, we propose to reduce the intrinsic style bias of CNNs to close the gap between domains. Our Style-Agnostic Networks (SagNets) disentangle style encodings from class categories to prevent style biased predictions and focus more on the contents. Extensive experiments show that our method effectively reduces the style bias and makes the model more robust under domain shift. It achieves remarkable performance improvements in a wide range of cross-domain tasks including domain generalization, unsupervised domain adaptation, and semi-supervised domain adaptation on multiple datasets.

Seokju Lee, Junsik Kim, Tae-Hyun Oh et al.

Dashboard cameras capture a tremendous amount of driving scene video each day. These videos are purposefully coupled with vehicle sensing data, such as from the speedometer and inertial sensors, providing an additional sensing modality for free. In this work, we leverage the large-scale unlabeled yet naturally paired data for visual representation learning in the driving scenario. A representation is learned in an end-to-end self-supervised framework for predicting dense optical flow from a single frame with paired sensing data. We postulate that success on this task requires the network to learn semantic and geometric knowledge in the ego-centric view. For example, forecasting a future view to be seen from a moving vehicle requires an understanding of scene depth, scale, and movement of objects. We demonstrate that our learned representation can benefit other tasks that require detailed scene understanding and outperforms competing unsupervised representations on semantic segmentation.

Inwan Yoo, Donggeun Yoo, Kyunghyun Paeng

Nuclei segmentation is one of the important tasks for whole slide image analysis in digital pathology. With the drastic advance of deep learning, recent deep networks have demonstrated successful performance of the nuclei segmentation task. However, a major bottleneck to achieving good performance is the cost for annotation. A large network requires a large number of segmentation masks, and this annotation task is given to pathologists, not the public. In this paper, we propose a weakly supervised nuclei segmentation method, which requires only point annotations for training. This method can scale to large training set as marking a point of a nucleus is much cheaper than the fine segmentation mask. To this end, we introduce a novel auxiliary network, called PseudoEdgeNet, which guides the segmentation network to recognize nuclei edges even without edge annotations. We evaluate our method with two public datasets, and the results demonstrate that the method consistently outperforms other weakly supervised methods.

Donggeun Yoo, In So Kweon

The performance of deep neural networks improves with more annotated data. The problem is that the budget for annotation is limited. One solution to this is active learning, where a model asks human to annotate data that it perceived as uncertain. A variety of recent methods have been proposed to apply active learning to deep networks but most of them are either designed specific for their target tasks or computationally inefficient for large networks. In this paper, we propose a novel active learning method that is simple but task-agnostic, and works efficiently with the deep networks. We attach a small parametric module, named "loss prediction module," to a target network, and learn it to predict target losses of unlabeled inputs. Then, this module can suggest data that the target model is likely to produce a wrong prediction. This method is task-agnostic as networks are learned from a single loss regardless of target tasks. We rigorously validate our method through image classification, object detection, and human pose estimation, with the recent network architectures. The results demonstrate that our method consistently outperforms the previous methods over the tasks.

Jongchan Park, Joon-Young Lee, Donggeun Yoo et al.

Learning-based color enhancement approaches typically learn to map from input images to retouched images. Most of existing methods require expensive pairs of input-retouched images or produce results in a non-interpretable way. In this paper, we present a deep reinforcement learning (DRL) based method for color enhancement to explicitly model the step-wise nature of human retouching process. We cast a color enhancement process as a Markov Decision Process where actions are defined as global color adjustment operations. Then we train our agent to learn the optimal global enhancement sequence of the actions. In addition, we present a 'distort-and-recover' training scheme which only requires high-quality reference images for training instead of input and retouched image pairs. Given high-quality reference images, we distort the images' color distribution and form distorted-reference image pairs for training. Through extensive experiments, we show that our method produces decent enhancement results and our DRL approach is more suitable for the 'distort-and-recover' training scheme than previous supervised approaches. Supplementary material and code are available at https://sites.google.com/view/distort-and-recover/

Dahun Kim, Donghyeon Cho, Donggeun Yoo et al.

In this paper, we explore methods of complicating self-supervised tasks for representation learning. That is, we do severe damage to data and encourage a network to recover them. First, we complicate each of three powerful self-supervised task candidates: jigsaw puzzle, inpainting, and colorization. In addition, we introduce a novel complicated self-supervised task called "Completing damaged jigsaw puzzles" which is puzzles with one piece missing and the other pieces without color. We train a convolutional neural network not only to solve the puzzles, but also generate the missing content and colorize the puzzles. The recovery of the aforementioned damage pushes the network to obtain robust and general-purpose representations. We demonstrate that complicating the self-supervised tasks improves their original versions and that our final task learns more robust and transferable representations compared to the previous methods, as well as the simple combination of our candidate tasks. Our approach achieves state-of-the-art performance in transfer learning on PASCAL classification and semantic segmentation.

Dahun Kim, Donghyeon Cho, Donggeun Yoo et al.

Weakly supervised semantic segmentation and localiza- tion have a problem of focusing only on the most important parts of an image since they use only image-level annota- tions. In this paper, we solve this problem fundamentally via two-phase learning. Our networks are trained in two steps. In the first step, a conventional fully convolutional network (FCN) is trained to find the most discriminative parts of an image. In the second step, the activations on the most salient parts are suppressed by inference conditional feedback, and then the second learning is performed to find the area of the next most important parts. By combining the activations of both phases, the entire portion of the tar- get object can be captured. Our proposed training scheme is novel and can be utilized in well-designed techniques for weakly supervised semantic segmentation, salient region detection, and object location prediction. Detailed experi- ments demonstrate the effectiveness of our two-phase learn- ing in each task.

Donggeun Yoo, Sunggyun Park, Kyunghyun Paeng et al.

A dominant paradigm for deep learning based object detection relies on a "bottom-up" approach using "passive" scoring of class agnostic proposals. These approaches are efficient but lack of holistic analysis of scene-level context. In this paper, we present an "action-driven" detection mechanism using our "top-down" visual attention model. We localize an object by taking sequential actions that the attention model provides. The attention model conditioned with an image region provides required actions to get closer toward a target object. An action at each time step is weak itself but an ensemble of the sequential actions makes a bounding-box accurately converge to a target object boundary. This attention model we call AttentionNet is composed of a convolutional neural network. During our whole detection procedure, we only utilize the actions from a single AttentionNet without any modules for object proposals nor post bounding-box regression. We evaluate our top-down detection mechanism over the PASCAL VOC series and ILSVRC CLS-LOC dataset, and achieve state-of-the-art performances compared to the major bottom-up detection methods. In particular, our detection mechanism shows a strong advantage in elaborate localization by outperforming Faster R-CNN with a margin of +7.1% over PASCAL VOC 2007 when we increase the IoU threshold for positive detection to 0.7.

Donggeun Yoo, Namil Kim, Sunggyun Park et al.

We present an image-conditional image generation model. The model transfers an input domain to a target domain in semantic level, and generates the target image in pixel level. To generate realistic target images, we employ the real/fake-discriminator as in Generative Adversarial Nets, but also introduce a novel domain-discriminator to make the generated image relevant to the input image. We verify our model through a challenging task of generating a piece of clothing from an input image of a dressed person. We present a high quality clothing dataset containing the two domains, and succeed in demonstrating decent results.

Donggeun Yoo, Sunggyun Park, Joon-Young Lee et al.

We present a novel detection method using a deep convolutional neural network (CNN), named AttentionNet. We cast an object detection problem as an iterative classification problem, which is the most suitable form of a CNN. AttentionNet provides quantized weak directions pointing a target object and the ensemble of iterative predictions from AttentionNet converges to an accurate object boundary box. Since AttentionNet is a unified network for object detection, it detects objects without any separated models from the object proposal to the post bounding-box regression. We evaluate AttentionNet by a human detection task and achieve the state-of-the-art performance of 65% (AP) on PASCAL VOC 2007/2012 with an 8-layered architecture only.

Donggeun Yoo, Sunggyun Park, Joon-Young Lee et al.

Compared to image representation based on low-level local descriptors, deep neural activations of Convolutional Neural Networks (CNNs) are richer in mid-level representation, but poorer in geometric invariance properties. In this paper, we present a straightforward framework for better image representation by combining the two approaches. To take advantages of both representations, we propose an efficient method to extract a fair amount of multi-scale dense local activations from a pre-trained CNN. We then aggregate the activations by Fisher kernel framework, which has been modified with a simple scale-wise normalization essential to make it suitable for CNN activations. Replacing the direct use of a single activation vector with our representation demonstrates significant performance improvements: +17.76 (Acc.) on MIT Indoor 67 and +7.18 (mAP) on PASCAL VOC 2007. The results suggest that our proposal can be used as a primary image representation for better performances in visual recognition tasks.