Shifting Weights: Adapting Object Detectors from Image to Video

By Kevin D. Tang, Vignesh Ramanathan, Li Fei-Fei, Daphne Koller. 2012

211 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Recognition in Terra Incognita (2018, methodology): “[48,49,50,51,52] look at methods for adapting detectors from one data source to another, such as from synthetic to real data or from images to video.”
Computational Baby Learning (2014, methodology): “learning pipeline is self-paced learning [18] which learns ﬁrst from easy samples and then from complex ones. Various vision applications based on self-paced learning have been proposed very recently [28] [16]. For example, Tang et al. [28] adapted object detectors learned from images to videos by s”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “ideos to obtain prospective positive target examples (windows) either by background modeling/subtraction [201, 202], or by combination of object tracking with regions proposed by the generic detector [203, 204, 205, 206] (see the main idea in Figure 25). Using these designated target samples in the ”
Discovering states and transformations in image collections (2015, background): “Similarly to previous works on transfer learning [6, 4, 19, 28] , our underlying assumption is the transferrability of knowledge between adjectives (state and transformation) (see Fig 2).”
Domain Adaptive Faster R-CNN for Object Detection in the Wild (2018, background): “There also exists work to learn detectors from alternative sources, such as from images to videos [54], from 3D models [46, 53], or from synthetic models [25].”
GANcraft: Unsupervised 3D Neural Rendering of Minecraft Worlds (2021, background): “3 [61] Kevin Tang, Vignesh Ramanathan, Li Fei-Fei, and Daphne Koller.”
A similarity measure for analyzing human activities using human-object interaction context (2014, methodology): “Although large labeled-image datasets are available, such as ImageNet [12] and Pascal VOC [13], it has been shown that detectors trained with still images have limited performance in dealing with video data [14].”
Unsupervised Hard Example Mining from Videos for Improved Object Detection (2018, unknown): “[54] adapt still-image object detectors to video by selecting training samples from unlabeled videos, based on the consistency between detections and tracklets, and then follow an iterative procedure that selects the easy examples from videos and hard examples from images to re-train the detector.”
Easy Samples First: Self-paced Reranking for Zero-Example Multimedia Search (2014, methodology): “This theory has been successfully applied to various applications, including domain adaptation [30], dictionary learning [31], segmentation [20], tracking [29], etc. Existing methods all adopt a simple binary weighting to obtain the global optimum for subproblems.”
Consistent video depth estimation (2020, methodology): “2008], adapting object detectors from images to videos [Jain and Learned-Miller 2011; Tang et al. 2012], and learning video-specific features for person re-identification [Cinbis et al.”
Easy Questions First? A Case Study on Curriculum Learning for Question Answering (2016, methodology): “The techniques have been shown to be useful in some computer vision tasks (Lee and Grauman, 2011; Kumar et al., 2011; Tang et al., 2012; Supancic and Ramanan, 2013; Jiang et al., 2014a).”
ConceptLearner: Discovering visual concepts from weakly labeled image collections (2014, methodology): “This is related to the topic of domain adaptation [28, 32], but we do not use the instances in the target domain to fine-tune the learned detectors.”
What Objective Does Self-paced Learning Indeed Optimize? (2015, background): “where Ω1 = O([−7,−8], 4)∩{[x, y]|y < x},Ω2 = O([6, 8], 4)∩{[x, y]|y > x},Ω3 = {x, y|x, y ∈ [−180, 180]}, O(x0, r) represents a circle area with center x0 and radius r, N(μ, σ(2)) denotes the Gaussian distribution with mean μ and variance σ(2) and U(Ω) represents the uniform distribution on Ω.”
Expanding object detector's Horizon: Incremental learning framework for object detection in videos (2015, methodology): “In [31] authors use the value of the loss (or margin) as confidence.”
From Red Wine to Red Tomato: Composition with Context (2017, background): “Another area related to our work is that of transfer learning [7–9, 45, 47, 56], feature embeddings [30, 52] and lowshot recognition [14, 16, 17, 58].”
Lucid Data Dreaming for Object Tracking (2017, background): “Ideally training data should be as similar as possible to the test data, even subtle differences may affect quality (e.g. training on static images for testing on videos underperforms [50]).”
STC: A Simple to Complex Framework for Weakly-Supervised Semantic Segmentation (2015, background): “Recently, various computer vision applications [31],”