{"url":"/method/sfam","slug":"sfam","name":"SFAM","full_name":"Scale-wise Feature Aggregation Module","full_name_withheld":false,"description_markdown":"**SFAM**, or **Scale-wise Feature Aggregation Module**, is a feature extraction block from the [M2Det](https://paperswithcode.com/method/m2det) architecture. It aims to aggregate the multi-level multi-scale features generated by [Thinned U-Shaped Modules](https://paperswithcode.com/method/tum) into a multi-level feature pyramid. \r\n\r\nThe first stage of SFAM is to concatenate features of the equivalent scale together along the channel dimension. The aggregated feature pyramid can be presented as $\\mathbf{X} =[\\mathbf{X}\\_1,\\mathbf{X}\\_2,\\dots,\\mathbf{X}\\_i]$, where $\\mathbf{X}\\_i = \\text{Concat}(\\mathbf{x}\\_i^1,\\mathbf{x}\\_i^2,\\dots,\\mathbf{x}\\_i^L) \\in \\mathbb{R}^{W\\_{i}\\times H\\_{i}\\times C}$ refers to the features of the $i$-th largest scale. Here, each scale in the aggregated pyramid contains features from multi-level depths. \r\n\r\nHowever, simple concatenation operations are not adaptive enough. In the second stage, we introduce a channel-wise attention module to encourage features to focus on channels that they benefit most. Following Squeeze-and-Excitation, we use [global average pooling](https://paperswithcode.com/method/global-average-pooling) to generate channel-wise statistics $\\mathbf{z} \\in \\mathbb{R}^C$ at the squeeze step. And to fully capture channel-wise dependencies, the following excitation step learns the attention mechanism via two fully connected layers:\r\n\r\n$$\r\n\\mathbf{s} = \\mathbf{F}\\_{ex}(\\mathbf{z},\\mathbf{W}) = \\sigma(\\mathbf{W}\\_{2} \\delta(\\mathbf{W}\\_{1}\\mathbf{z})),\r\n$$\r\n\r\nwhere $\\sigma$ refers to the [ReLU](https://paperswithcode.com/method/relu) function, $\\delta$ refers to the sigmoid function, $\\mathbf{W}\\_{1} \\in \\mathbb{R}^{\\frac{C}{r}\\times C}$ , $\\mathbf{W}\\_{2} \\in \\mathbb{R}^{C\\times \\frac{C}{r}}$, r is the reduction ratio ($r=16$ in our experiments). The final output is obtained by reweighting the input $\\mathbf{X}$ with activation $\\mathbf{s}$:\r\n\r\n$$\r\n\\tilde{\\mathbf{X}}_i^c = \\mathbf{F}\\_{scale}(\\mathbf{X}\\_i^c,s_c) = s_c \\cdot \\mathbf{X}_i^c,\r\n$$\r\n\r\nwhere $\\tilde{\\mathbf{X}\\_i} = [\\tilde{\\mathbf{X}}\\_i^1,\\tilde{\\mathbf{X}}\\_i^2,...,\\tilde{\\mathbf{X}}\\_i^C]$, each of the features is enhanced or weakened by the rescaling operation.","description_state":"present","introduced_year":null,"introduced_by":{"title":null,"paper":null,"first_author":null,"n_authors":0,"url_abs":null,"archive_paper_url":null},"source":{"url":"http://arxiv.org/abs/1811.04533v3","title":"M2Det: A Single-Shot Object Detector based on Multi-Level Feature Pyramid Network","url_on_a_paper_host":true},"code_snippet_url":"https://github.com/qijiezhao/M2Det/blob/ade4f3d12979800c367bf1e46d2e316e73a87514/layers/nn_utils.py#L133","code_snippet_url_on_a_code_host":true,"categories":[{"area":"Computer Vision","area_id":"computer-vision","collection":"Feature Extractors","url":"/methods/category/feature-extractors","pwc_aliases":[]}],"n_papers_tagged":5,"archive_num_papers":null,"papers_newest_first":[{"paper":"/paper/emphasizing-crucial-features-for-efficient","title":"Emphasizing Crucial Features for Efficient Image Restoration","date":"2024-05-19","arxiv_id":"2405.11468","n_code_links":1,"syntology":null},{"paper":"/paper/towards-robust-visual-information-extraction","title":"Towards Robust Visual Information Extraction in Real World: New Dataset and Novel Solution","date":"2021-01-24","arxiv_id":"2102.06732","n_code_links":1,"syntology":null},{"paper":null,"title":"Structure from Articulated Motion: Accurate and Stable Monocular 3D Reconstruction without Training Data","date":"2019-05-12","arxiv_id":"1905.04789","n_code_links":0,"syntology":null},{"paper":null,"title":"Adversarial Learning-Based On-Line Anomaly Monitoring for Assured Autonomy","date":"2018-11-12","arxiv_id":"1811.04539","n_code_links":0,"syntology":null},{"paper":"/paper/m2det-a-single-shot-object-detector-based-on","title":"M2Det: A Single-Shot Object Detector based on Multi-Level Feature Pyramid Network","date":"2018-11-12","arxiv_id":"1811.04533","n_code_links":11,"syntology":{"ran":1,"of":1,"unverified":0,"pointer_only":0}}],"papers_shown":5,"tasks":[{"task":"/task/3d-feature-matching","name":"3D Feature Matching","papers":1},{"task":"/task/3d-reconstruction","name":"3D Reconstruction","papers":1},{"task":"/task/anomaly-detection","name":"Anomaly Detection","papers":1},{"task":"/task/decoder","name":"Decoder","papers":1},{"task":null,"name":"Generative Adversarial Network","papers":1},{"task":"/task/image-restoration","name":"Image Restoration","papers":1},{"task":"/task/object","name":"Object","papers":1},{"task":"/task/object-detection","name":"Object Detection","papers":1},{"task":"/task/text-detection","name":"Text Detection","papers":1},{"task":"/task/text-spotting","name":"Text Spotting","papers":1},{"task":"/task/video-prediction","name":"Video Prediction","papers":1},{"task":"/task/document-understanding","name":"document understanding","papers":1},{"task":"/task/object-detection-1","name":"object-detection","papers":1}],"tasks_shown":13,"n_tasks":13,"usage_by_year":[{"year":"2018","papers":2},{"year":"2019","papers":1},{"year":"2021","papers":1},{"year":"2024","papers":1}],"row_source":"embedded","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/sfam"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}