{"url":"/method/scale-aggregation-block","slug":"scale-aggregation-block","name":"Scale Aggregation Block","full_name":"Scale Aggregation Block","full_name_withheld":false,"description_markdown":"A **Scale Aggregation Block** concatenates feature maps at a wide range of scales. Feature maps for each scale are generated by a stack of downsampling, [convolution](https://paperswithcode.com/method/convolution) and upsampling operations. The proposed scale aggregation block is a standard computational module which readily replaces any given transformation $\\mathbf{Y}=\\mathbf{T}(\\mathbf{X})$, where $\\mathbf{X}\\in \\mathbb{R}^{H\\times W\\times C}$, $\\mathbf{Y}\\in \\mathbb{R}^{H\\times W\\times C_o}$ with $C$ and $C_o$ being the input and output channel number respectively. $\\mathbf{T}$ is any operator such as a convolution layer or a series of convolution layers. Assume we have $L$ scales. Each scale $l$ is generated by sequentially conducting a downsampling $\\mathbf{D}_l$, a transformation $\\mathbf{T}_l$ and an unsampling operator $\\mathbf{U}_l$:\r\n\r\n$$\r\n\\mathbf{X}^{'}_l=\\mathbf{D}_l(\\mathbf{X}),\r\n\\label{eq:eq_d}\r\n$$\r\n\r\n$$\r\n\\mathbf{Y}^{'}_l=\\mathbf{T}_l(\\mathbf{X}^{'}_l),\r\n\\label{eq:eq_tl}\r\n$$\r\n\r\n$$\r\n\\mathbf{Y}_l=\\mathbf{U}_l(\\mathbf{Y}^{'}_l),\r\n\\label{eq:eq_u}\r\n$$\r\n\r\nwhere $\\mathbf{X}^{'}_l\\in \\mathbb{R}^{H_l\\times W_l\\times C}$,\r\n$\\mathbf{Y}^{'}_l\\in \\mathbb{R}^{H_l\\times W_l\\times C_l}$, and\r\n$\\mathbf{Y}_l\\in \\mathbb{R}^{H\\times W\\times C_l}$.\r\nNotably, $\\mathbf{T}_l$ has the similar structure as $\\mathbf{T}$.\r\nWe can concatenate all $L$ scales together, getting\r\n\r\n$$\r\n\\mathbf{Y}^{'}=\\Vert^L_1\\mathbf{U}_l(\\mathbf{T}_l(\\mathbf{D}_l(\\mathbf{X}))),\r\n\\label{eq:eq_all}\r\n$$\r\n\r\nwhere $\\Vert$ indicates concatenating feature maps along the channel dimension, and $\\mathbf{Y}^{'} \\in \\mathbb{R}^{H\\times W\\times \\sum^L_1 C_l}$ is the final output feature maps of the scale aggregation block.\r\n\r\nIn the reference implementation, the downsampling $\\mathbf{D}_l$ with factor $s$ is implemented by a max pool layer with $s\\times s$ kernel size and  $s$ stride. The upsampling $\\mathbf{U}_l$ is implemented by resizing with the nearest neighbor  interpolation.","description_state":"present","introduced_year":null,"introduced_by":{"title":"Data-Driven Neuron Allocation for Scale Aggregation Networks","paper":"/paper/190409460","first_author":"Yi Li","n_authors":4,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/190409460"},"source":{"url":"http://arxiv.org/abs/1904.09460v1","title":"Data-Driven Neuron Allocation for Scale Aggregation Networks","url_on_a_paper_host":true},"code_snippet_url":"https://github.com/Eli-YiLi/ScaleNet/blob/2c27b4207691dbe72f7e19fd88bfccc5ce5b3080/pytorch/scalenet.py#L8","code_snippet_url_on_a_code_host":true,"categories":[{"area":"Computer Vision","area_id":"computer-vision","collection":"Image Model Blocks","url":"/methods/category/image-model-blocks","pwc_aliases":[]}],"n_papers_tagged":7,"archive_num_papers":7,"papers_newest_first":[{"paper":"/paper/scale-invariance-of-graph-neural-networks","title":"Scale Invariance of Graph Neural Networks","date":"2024-11-28","arxiv_id":"2411.19392","n_code_links":1,"syntology":null},{"paper":"/paper/scalenet-scale-invariance-learning-in","title":"ScaleNet: Scale Invariance Learning in Directed Graphs","date":"2024-11-13","arxiv_id":"2411.08758","n_code_links":1,"syntology":null},{"paper":null,"title":"ScaleNet: An Unsupervised Representation Learning Method for Limited Information","date":"2023-10-03","arxiv_id":"2310.02386","n_code_links":0,"syntology":null},{"paper":"/paper/scalenet-searching-for-the-model-to-scale","title":"ScaleNet: Searching for the Model to Scale","date":"2022-07-15","arxiv_id":"2207.07267","n_code_links":1,"syntology":{"ran":0,"of":4,"unverified":4,"pointer_only":0}},{"paper":"/paper/scalenet-a-shallow-architecture-for-scale","title":"ScaleNet: A Shallow Architecture for Scale Estimation","date":"2021-12-09","arxiv_id":"2112.04846","n_code_links":1,"syntology":null},{"paper":"/paper/scalenas-one-shot-learning-of-scale-aware","title":"ScaleNAS: One-Shot Learning of Scale-Aware Representations for Visual Recognition","date":"2020-11-30","arxiv_id":"2011.14584","n_code_links":0,"syntology":null},{"paper":"/paper/190409460","title":"Data-Driven Neuron Allocation for Scale Aggregation Networks","date":"2019-04-20","arxiv_id":"1904.09460","n_code_links":1,"syntology":null}],"papers_shown":7,"tasks":[{"task":"/task/graph-learning","name":"Graph Learning","papers":2},{"task":"/task/image-classification","name":"Image Classification","papers":2},{"task":"/task/node-classification","name":"Node Classification","papers":2},{"task":"/task/pose-estimation","name":"Pose Estimation","papers":2},{"task":"/task/3d-reconstruction","name":"3D Reconstruction","papers":1},{"task":"/task/camera-pose-estimation","name":"Camera Pose Estimation","papers":1},{"task":"/task/classification-1","name":"Classification","papers":1},{"task":"/task/geometric-matching","name":"Geometric Matching","papers":1},{"task":"/task/multi-person-pose-estimation","name":"Multi-Person Pose Estimation","papers":1},{"task":"/task/architecture-search","name":"Neural Architecture Search","papers":1},{"task":"/task/node-classification-on-non-homophilic","name":"Node Classification on Non-Homophilic (Heterophilic) Graphs","papers":1},{"task":"/task/object-detection","name":"Object Detection","papers":1},{"task":"/task/one-shot-learning","name":"One-Shot Learning","papers":1},{"task":"/task/representation-learning","name":"Representation Learning","papers":1},{"task":"/task/semantic-segmentation","name":"Semantic Segmentation","papers":1},{"task":"/task/image-classification","name":"image-classification","papers":1},{"task":"/task/model","name":"model","papers":1},{"task":"/task/object-detection-1","name":"object-detection","papers":1}],"tasks_shown":18,"n_tasks":18,"usage_by_year":[{"year":"2019","papers":1},{"year":"2020","papers":1},{"year":"2021","papers":1},{"year":"2022","papers":1},{"year":"2023","papers":1},{"year":"2024","papers":2}],"row_source":"methods_table","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/scale-aggregation-block"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}