Beyond Comparing Image Pairs: Setwise Active Learning for Relative Attributes

Liang, Lucy; Grauman, Kristen

doi:10.1109/cvpr.2014.34

Cited by 45 publications

(24 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While, the heuristic we propose in this paper is formulated based on similar principles, the actual implementation differs vastly not only due to difference in dataset but also due to a difference in approach. The current literature contains various papers that describe learning as "setwise" [11], but the setwise learning used in this paper is strictly as described in [5] and [4] and is described in further details in Section III. Our analysis proceeds as a two step inspection:…”

Section: Related Workmentioning

confidence: 99%

Calculating Patient Similarity Based on Respiration Induced Tumor Motion

Shamsuddin

Balasubramanian

Sawant

et al. 2015

2015 International Conference on Healthcare Informatics

View full text Add to dashboard Cite

In stereotactic radiotherapy for thoracic and abdominal tumors, respiratory motion management is crucial for improving efficacy of treatment, while minimizing risk to heathy tissue and organs. Since tumor motion exhibits dynamic variation in characteristics, analyzing the behavior distribution of tumor motion can improve treatment planning. Identifying similarities among patients is an intuitive step towards patient profiling for optimal treatment strategy. This paper investigates how patients, undergoing radiotherapy, can be clustered into five treatment groups, which are identified by medical physicists. And in order to do so, we propose a combination of an adaptive segmentation technique and couple it with a setwise clustering approach for measuring similarity between patients. Our methods aim to group the patients solely based on their tumor motion traces brought about by respiration. Through a series of empirical and statistical analyses, we explore the impact of different feature sets and processing methodologies on the proposed patient grouping approach.

show abstract

Section: Related Workmentioning

confidence: 99%

Calculating Patient Similarity Based on Respiration Induced Tumor Motion

Shamsuddin

Balasubramanian

Sawant

et al. 2015

2015 International Conference on Healthcare Informatics

View full text Add to dashboard Cite

show abstract

“…For visual recognition problems, pool-based methods are the norm: the learner scans a pool of unlabeled samples and iteratively queries for the label on one or more of them based on uncertainty or expected model influence (e.g., [6], [7], [8], [5]). Active ranking models adapt the concepts of pool-based learning to select pairs for comparison [54], [55]. Hard negative mining-often used for training object detectors [56], [57]-also focuses the learner's attention on useful samples, though in this case from a pool of alreadylabeled data.…”

Section: Related Workmentioning

confidence: 99%

Thinking Outside the Pool: Active Training Image Creation for Relative Attributes

Grauman

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Self Cite

View full text Add to dashboard Cite

Current wisdom suggests more labeled image data is always better, and obtaining labels is the bottleneck. Yet curating a pool of sufficiently diverse and informative images is itself a challenge. In particular, training image curation is problematic for fine-grained attributes, where the subtle visual differences of interest may be rare within traditional image sources. We propose an active image generation approach to address this issue. The main idea is to jointly learn the attribute ranking task while also learning to generate novel realistic image samples that will benefit that task. We introduce an end-to-end framework that dynamically "imagines" image pairs that would confuse the current model, presents them to human annotators for labeling, then improves the predictive model with the new examples. With results on two datasets, we show that by thinking outside the pool of real images, our approach gains generalization accuracy for challenging fine-grained attribute comparisons.

show abstract

“…To solve this problem, they alternate between fixing the model parameters and the set of images to label. In the context of relative attribute annotations and learning relative attribute models, Liang and Grauman [2014] show that asking humans to fully order sets of 4 images, rather than to provide annotations on pairs of images, allows the system to learn relative attribute models faster. Further, the cost of obtaining the full ordering on the 4 images is about the same as on ordering just 2 images.…”

Section: Practical Concerns and Selecting Batches Of Labelsmentioning

confidence: 99%