Noname manuscript No. (will be inserted by the editor) A Numerical Study of the Bottom-up and Top-down Inference Processes in And-Or Graphs

Citations in the corpus

• I2T: Image Parsing to Text Description (2010, methodology): “Here we extend the discussion with several improvements, namely, a bottom–up/top–down inference algorithm from [64] and a cluster sampling algorithm from [51].”
• Towards Interpretable Object Detection by Unfolding Latent Structures (2019, methodology): “…could unfold the space of latent part configurations, which is usually huge, we can evaluate the interpretability score in the spirit similar to the masking and scaling operators used in [61] for evaluating information contributions of bottom-up/top-down computing processes in a hierarchical model.”
• Top-Down Learning for Structured Labeling with Convolutional Pseudoprior (2015, background): “Systems trying to explicitly incorporate topdown information under the Bayesian formulation point to a promising direction [15,16,17,18] but in the absence of a clear solution.”
• Interpretable R-CNN (2017, methodology): “If we could unfold the space of latent part configurations, which is usually huge, we can evaluate the explainability score in the spirit similar to the masking and scaling operators used in [75] for evaluating information contributions of bottom-up/top-down computing processes in a hierarchical mod”
• A generative vision model that trains with high data efficiency and breaks text-based CAPTCHAs (2017, methodology): “Con-tours are modeled using a compositional hierarchy of features ( 28 , 39 ).”
• DeepContext: Context-Encoding Neural Pathways for 3D Holistic Scene Understanding (2016, background): “The role of context has been studied extensively in computer vision [27, 41, 5, 4, 3, 7, 18, 37, 33, 35, 34, 19, 12, 9, 43, 1, 39, 24, 10, 42, 20, 21].”
• Joint Deep Learning for Pedestrian Detection (2013, methodology): “Some of them use the detection scores of blocks or parts [51, 34, 13, 54] as input for visibility estimation.”
• A discriminative deep model for pedestrian detection with occlusion handling (2012, background): “These approaches obtain the h̃i in (3) from thresholding detection scores [29, 33], from linear SVM in [10] when only intensity information is available or from other cues like depth and motion [10].”
• From Red Wine to Red Tomato: Composition with Context (2017, background): “The parts-based systems such as Deformable Part Models [15], grammars [25, 44, 59, 62, 67, 69], and AND-OR graphs [54, 61, 66] also rely on compositionality of objects to build recognition systems.”