Learning and example selection for object and pattern detection

By Kah Kay Sung, Tomaso A. Poggio. 1996

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Natural Adversarial Examples (2019, methodology): “Like this recent concurrent research, we also use adversarial filtration [48], but the technique of adversarial filtration has not been applied to collecting image datasets until this paper.”
Training Region-Based Object Detectors with Online Hard Example Mining (2016, methodology): “The idea of dataset boot-strapping [25, 30], typically called hard negative mining in recent work [12], appears in the training of most successful object detectors [7, 9, 12, 13, 15, 16, 23, 25, 29].”
Focal Loss for Dense Object Detection (2017, background): “A common solution is to perform some form of hard negative mining [9], [16], [17], [19], [20] that samples hard examples during training or more complex sampling/”
VoxCeleb2: Deep Speaker Recognition (2018, methodology): “We use an ofﬂine hard negative mining strategy, which allows us to select harder negatives ( e.g. top 1-percent of randomly generated pairs) than is possible with online (in-batch) hard negative mining methods [42, 41, 43] limited by the batch size.”
Reliable Uncertainty Estimates in Deep Neural Networks using Noise Contrastive Priors (2018, neutral): “Error reduction approaches incorporate the select data based on the predicted reduction in classiﬁer error based on information (MacKay, 1992a), Monte Carlo estimation (Roy and McCallum, 2001), or hard-negative example mining (Sung, 1994; Rowley et al., 1998).”
Supervision via competition: Robot adversaries for learning tasks (2016, background): “[5], [6], formerly referred to as bootstrap learning.”
Example-Based Learning for View-Based Human Face Detection (1998, positive): “We stress again that the underlying technique is fairly general and has been recently used by ourselves and others for taking on feature-detection and pattern-recognition tasks in other problem domains [11], [ 17 ].”
Mining on Manifolds: Metric Learning Without Labels (2018, background): “This principle has been known much earlier as bootstrapping [54].”
Localizing and segmenting text in images and videos (2002, unknown): “This iterative box expansion repeats until the text box stops growing.”
Voxceleb: Large-scale speaker verification in the wild (2020, methodology): “We use an of ﬂ ine hard negative mining strategy, which allows us to select harder negatives ( e.g. top 1-percent of randomly generated pairs) than is possible with online (in-batch) hard negative mining methods (Sung, 1996; Hermans et al., 2017; Song et al., 2016) limited by the batch size.”
Learnable PINs: Cross-Modal Embeddings for Person Identity (2018, background): “In such a scenario, the network rapidly learns to correctly map the easy examples, but hard positive and negative mining [11, 18, 36, 38, 42] is often required to improve performance further by e ciently sampling examples”
Neural network-based face detection (1996, background): “A typical training run selects approximately 8000 non-face images from the 146,212,178 subimages that are available at all locations and scales in the training scenery images.”
A Trainable System for Object Detection (2000, unknown): “These images are all scaled to the dimensions 19£ 19 and show the face from above the eyebrows to below the lips; typical images from the database are shown in Fig. 3. Databases of this size and composition have been used extensively in face detection ( Sung, 1995; Rowley et al., 1998; Osuna et al.”
Automatic Adaptation of Object Detectors to New Domains Using Self-Training (2019, background): “We expand the set of pseudo-labels to include these “tracker-only” boundingboxes that were missed by the baseline detector, since these hard examples are expected to have a larger influence on the model’s decision boundary upon retraining [52, 48, 25].”