{ "id": "ep0cyh1hnt", "slug": "fully-convolutional-network", "title": "Fully convolutional network", "updated": "2026-10-15", "topic_path": [ { "id": "technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/topics/technology" }, { "id": "technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing" }, { "id": "technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai" }, { "id": "technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.ml_computation" }, { "id": "technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep" }, { "id": "technology.computing.ai.neural_deep.alpha-a-to-h", "label": "Titles A to H", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep.alpha-a-to-h" } ], "geo": [ { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Non-geographic · 2001 to 2020: Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep", "path": [ { "id": "geo.nongeo", "label": "Non-geographic", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo" }, { "id": "geo.nongeo.t2001", "label": "Non-geographic · 2001 to 2020", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001" }, { "id": "geo.nongeo.t2001.technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology" }, { "id": "geo.nongeo.t2001.technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing" }, { "id": "geo.nongeo.t2001.technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai" }, { "id": "geo.nongeo.t2001.technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.ml_computation" }, { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep" } ] } ], "excerpt": "A fully convolutional network (FCN) is a neural network architecture for image segmentation that replaces fully connected layers with convolutions, producing dense per-pixel predictions from inputs of any size.", "snippet": "A fully convolutional network (FCN) is a neural network architecture for image segmentation that replaces fully connected layers with convolutions, producing dense per-pixel predictions from inputs of any size.", "node": "technology.computing.ai.neural_deep.alpha-a-to-h", "markdown": "# Fully convolutional network\n\nA fully convolutional network (FCN) is a neural network architecture for image segmentation that replaces a classifier's fully connected layers with convolutional layers, so the network produces a dense per-pixel prediction map from an input of any size. Introduced for semantic segmentation, the original FCN reached 62.2% mean intersection over union (mean IU) on [PASCAL VOC](https://www.edgepedia.org/pascal-voc) 2012, a 20% relative improvement at the time, with inference in under one fifth of a second per image.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) Because every layer is convolutional (or pointwise), the network keeps spatial information that fully connected layers discard and avoids the fixed input sizes those layers impose.[2](https://www.cs.ubc.ca/%7Eschmidtm/Courses/440-W21/L35.pdf)\n\n| Key fact | Value |\n|---|---|\n| Introducing paper | Shelhamer, Long, Darrell, IEEE TPAMI 2016[3](https://doi.org/10.1109/tpami.2016.2572683) |\n| Journal edition result | 67.2% mean IU on PASCAL VOC 2012, ~0.1 s inference[4](https://arxiv.org/html/1605.06211v1) |\n| Speedup over patchwise pipelines | 114× (convnet only) to 286× (overall) versus R-CNN/SDS-style approaches[5](https://arxiv.org/abs/1411.4038) |\n| Skip variants | FCN-32s, FCN-16s, FCN-8s, fusing progressively finer layers[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) |\n| Training loss | Per-pixel multinomial logistic loss; evaluation by mean pixel IoU[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) |\n| Main limitation | Coarse output from the prediction layer's pixel stride (32 pixels in FCN-32s)[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) |\n\n## How it works\n\nA classification network ends in fully connected layers that map a fixed-size feature volume to a class vector. Any fully connected layer is equivalent to a convolution whose kernels cover the entire input region; recasting the layers this way turns the classifier into a network that accepts input of any size and outputs a spatial map of class scores instead of a single vector.[5](https://arxiv.org/abs/1411.4038) Each output unit then corresponds to a receptive field in the input, so one forward pass yields a grid of predictions rather than one label.\n\nThe coarse grid must be upsampled to full resolution. Upsampling by a factor \\( f \\) is convolution with a fractional input stride of \\( 1/f \\), implemented as backwards convolution (often called deconvolution) with output stride \\( f \\), and its weights can be learned end-to-end by backpropagation from the pixelwise loss.[5](https://arxiv.org/abs/1411.4038) A transposed convolution with stride \\( s \\), padding \\( s/2 \\), and kernel size \\( 2s \\) increases the input's height and width by a factor of \\( s \\).[6](https://d2l.ai/chapter_computer-vision/fcn.html)\n\nDeep layers carry semantic information but at coarse resolution; shallow layers carry appearance detail. The FCN's skip architecture combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer, turning the network's line topology into a directed acyclic graph with edges skipping ahead from shallower to deeper layers.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) • [4](https://arxiv.org/html/1605.06211v1)\n\n## How it is done\n\n1. Start from a pretrained classifier. The original work adapted AlexNet, VGG, and [GoogLeNet](https://www.edgepedia.org/googlenet), and fine-tuned their learned representations for segmentation[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html); the reference implementation fine-tunes FCN-32s from the ILSVRC-trained VGG-16 model.[7](https://github.com/shelhamer/fcn.berkeleyvision.org/tree/master/)\n2. Convert the fully connected layers to convolutions and append a 1×1 convolution mapping channels to the number of classes (21 for PASCAL VOC 2012), followed by a transposed convolution that restores input resolution.[6](https://d2l.ai/chapter_computer-vision/fcn.html)\n3. Initialize upsampling layers to bilinear interpolation. In the original experiments the bilinear kernels were then learned; the reference implementation fixes the final layer's backward convolution weights to bilinear interpolation with no significant accuracy difference and a slight speed-up.[4](https://arxiv.org/html/1605.06211v1) • [7](https://github.com/shelhamer/fcn.berkeleyvision.org/tree/master/)\n4. Train with a per-pixel multinomial logistic loss and validate with mean pixel intersection over union, averaged over all classes including background, ignoring masked-out ambiguous pixels.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html)\n5. Build the skip nets in stages, dropping the learning rate 100× from FCN-32s to FCN-16s and 100× more from FCN-16s to FCN-8s, which the authors found necessary for continued improvement.[4](https://arxiv.org/html/1605.06211v1)\n\nInputs whose height or width is not divisible by 32 do not generally require cropping into 32-multiple rectangles and averaging overlapping upsampled outputs.[6](https://d2l.ai/chapter_computer-vision/fcn.html) • [1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html)\n\n## Origin\n\nThe FCN was introduced by Evan Shelhamer, Jonathan Long, and [Trevor Darrell](https://www.edgepedia.org/trevor-darrell) in \"Fully Convolutional Networks for Semantic Segmentation\", published in [IEEE Transactions on Pattern Analysis and Machine Intelligence](https://www.edgepedia.org/ieee-transactions-on-pattern-analysis-and-machine-intelligence) in 2016.[3](https://doi.org/10.1109/tpami.2016.2572683) An earlier conference version appeared at CVPR 2015.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) • [5](https://arxiv.org/abs/1411.4038) The paper states it is the first work to train FCNs end-to-end for pixelwise prediction from supervised pre-training.[5](https://arxiv.org/abs/1411.4038)\n\nThe paper credits precursors: Wolf and Platt expanded convnet outputs to 2D detection-score maps; and Ning and colleagues built a convnet for coarse multiclass segmentation of C. elegans tissues with fully convolutional inference.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) The input-shifting and output-interlacing trick for dense predictions from coarse outputs was introduced by OverFeat (Sermanet and colleagues, 2013), and the FCN paper offers deconvolution layers as an alternative.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) • [8](https://doi.org/10.48550/arxiv.1312.6229)\n\n## Variants\n\nThe variant names encode the final prediction stride. FCN-32s predicts from the stride-32 layer alone; FCN-16s fuses pool4 predictions, improving validation performance by 3.0 mean IU to 62.4 in the conference version; FCN-8s adds pool3.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) In the journal edition, skips are implemented by scoring each layer with a 1×1 convolution, aligning the scores by interpolation and cropping, and summing them; max fusion was rejected because learning was difficult due to gradient switching.[4](https://arxiv.org/html/1605.06211v1)\n\nFCN developments led to U-Net and its derivatives.[9](https://www.mdpi.com/2075-4418/12/11/2765) U-Net (Ronneberger, Fischer, and Brox, 2015) combines skip layers and learned deconvolution for pixel labeling of microscopy images[4](https://arxiv.org/html/1605.06211v1) • [10](https://doi.org/10.48550/arxiv.1505.04597); U-Net++ fills the space between encoder and decoder with dense convolutional blocks whose horizontal skips cross up to three blocks.[9](https://www.mdpi.com/2075-4418/12/11/2765) SegNet introduced max unpooling with saved indices, not present in the FCN paper.[11](https://tjmachinelearning.com/lectures/1718/fcn/fcn.pdf) DeepLab (Chen and colleagues, 2016) raises output resolution with atrous convolution and reinforces edges with a fully connected CRF; DeepLab v3 applies atrous convolution with ASPP, increasing mIOU to 70–80%.[9](https://www.mdpi.com/2075-4418/12/11/2765) • [12](https://doi.org/10.48550/arxiv.1606.00915) FC-DenseNet (Jégou and colleagues, 2016) extends DenseNets to segmentation with transition-up modules of 3×3 transposed convolutions with stride 2 plus skip connections.[13](https://doi.org/10.48550/arxiv.1611.09326) IFCN-8s adds dense skip connections from all feature maps starting with pool3, versus only pool3 and pool4 in FCN-8s, at minor extra computation cost.[14](https://ar5iv.labs.arxiv.org/html/1611.08986)\n\n## Applications\n\nBeyond PASCAL VOC, the original FCN reported state-of-the-art segmentation of NYUDv2 and SIFT-Flow.[1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) Fully convolutional inference had already been exploited for sliding-window detection by Sermanet and colleagues and semantic segmentation by Pinheiro and Collobert.[5](https://arxiv.org/abs/1411.4038) In medical imaging, the FCN lineage runs through U-Net and its derivatives[9](https://www.mdpi.com/2075-4418/12/11/2765), and the Fully Convolutional Transformer (Tragakis and colleagues, 2022) extends the FCN encoder–decoder paradigm by replacing self-attention projections with depthwise convolutions, removing the need for positional encoding.[15](https://doi.org/10.48550/arxiv.2206.00566) FC-DenseNet improved the state of the art on the Camvid and Gatech urban scene understanding benchmarks.[13](https://doi.org/10.48550/arxiv.1611.09326)\n\n## Limitations and alternatives\n\nThe dominant failure mode is coarse boundaries. Subsampling reduces the output from the input size by a factor equal to the pixel stride of the output units' receptive fields, and the 32-pixel stride of FCN-32s limits the detail in the upsampled output.[4](https://arxiv.org/html/1605.06211v1) • [1](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html) [Upsampling](https://www.edgepedia.org/upsampling) with transposed convolutions or unpooling loses information; skip connections from earlier, higher-resolution layers, deconvolutional networks, and combinations with CRFs are the standard remedies.[11](https://tjmachinelearning.com/lectures/1718/fcn/fcn.pdf) • [2](https://www.cs.ubc.ca/%7Eschmidtm/Courses/440-W21/L35.pdf) IFCN's authors add a domain gap: the pretrained CNN is trained on low-resolution 224×224 classification images while segmentation inputs are typically high resolution (for example 512×512), causing locally ambiguous predictions.[14](https://ar5iv.labs.arxiv.org/html/1611.08986)\n\nAgainst the sliding-window alternative, the gain is computational. Classifying each pixel from a neighborhood crop would require 40,000 forward passes for a 200×200 image.[2](https://www.cs.ubc.ca/%7Eschmidtm/Courses/440-W21/L35.pdf) Overall inference time fell 114× (convnet only) to 286× (overall) relative to R-CNN/SDS-style pipelines.[5](https://arxiv.org/abs/1411.4038)\n\nLater transformer-based methods compete on both axes. SegNeXt-S (Guo and colleagues, 2022) outperforms SegFormer-B2 on Cityscapes (81.3% vs 81.0% mIoU) using about 1/6 the compute (124.6G vs 717.1G).[16](https://doi.org/10.48550/arxiv.2209.08575) FCN-lineage systems remain active: IFCN-8s with VGG-16 beats FCN-8s by more than 7.5% mIOU on all tested benchmarks, reaching 74.6 mIOU without CRF and 75.3 with CRF on PASCAL VOC 2012[14](https://ar5iv.labs.arxiv.org/html/1611.08986), and Panoptic FCN (Li and colleagues, 2020) applies fully convolutional networks to panoptic segmentation.[17](https://doi.org/10.48550/arxiv.2012.00720)\n\n## References\n\n1. [Fully Convolutional Networks for Semantic Segmentation (CVPR 2015 open access page)](https://openaccess.thecvf.com/content_cvpr_2015/html/Long_Fully_Convolutional_Networks_2015_CVPR_paper.html)\n2. [CPSC 440 lecture notes: Fully-Convolutional Networks](https://www.cs.ubc.ca/%7Eschmidtm/Courses/440-W21/L35.pdf)\n3. [Evan Shelhamer, Jonathan Long, Trevor Darrell (2016). Fully Convolutional Networks for Semantic Segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence.](https://doi.org/10.1109/tpami.2016.2572683)\n4. [Fully Convolutional Networks for Semantic Segmentation (PAMI 2016 journal edition, arXiv:1605.06211)](https://arxiv.org/html/1605.06211v1)\n5. [Fully Convolutional Networks for Semantic Segmentation (arXiv:1411.4038)](https://arxiv.org/abs/1411.4038)\n6. [Dive into Deep Learning, Fully Convolutional Networks](https://d2l.ai/chapter_computer-vision/fcn.html)\n7. [shelhamer/fcn.berkeleyvision.org, reference implementation (GitHub)](https://github.com/shelhamer/fcn.berkeleyvision.org/tree/master/)\n8. [Sermanet, Pierre and colleagues (2013). OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1312.6229)\n9. [Fully Convolutional Network for the Semantic Segmentation of Medical Images: A Survey](https://www.mdpi.com/2075-4418/12/11/2765)\n10. [Ronneberger, Olaf, Fischer, Philipp, Brox, Thomas (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1505.04597)\n11. [Fully Convolutional Networks (Sardana, lecture notes)](https://tjmachinelearning.com/lectures/1718/fcn/fcn.pdf)\n12. [Chen, Liang-Chieh and colleagues (2016). DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1606.00915)\n13. [Jégou, Simon and colleagues (2016). The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1611.09326)\n14. [Improving Fully Convolution Network for Semantic Segmentation (IFCN)](https://ar5iv.labs.arxiv.org/html/1611.08986)\n15. [Tragakis, Athanasios and colleagues (2022). The Fully Convolutional Transformer for Medical Image Segmentation. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2206.00566)\n16. [Guo, Meng-Hao and colleagues (2022). SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2209.08575)\n17. [Li, Yanwei and colleagues (2020). Fully Convolutional Networks for Panoptic Segmentation. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2012.00720)\n\n---\n*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Artificial intelligence and data › Machine learning and neural computation › Neural networks and deep learning › Titles A to H*\n\n*Initially written Sep 29, 2026 · Edited: Oct 15, 2026*\n\n*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*\n\nLicense: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license\n", "same_as": [], "url": "https://www.edgepedia.org/fully-convolutional-network", "markdown_url": "https://www.edgepedia.org/fully-convolutional-network.md", "license": { "name": "Edgepedia Community License 1.0", "url": "https://www.edgepedia.org/edgepedia/license", "summary": "Free with credit, commercial use included. AI training is open to everyone. For other uses, organizations over USD 100M in revenue or 100M monthly users license separately.", "spdx": "LicenseRef-Edgepedia-Community-1.0" }, "credit": "\"Fully convolutional network\", Edgepedia (EdgeChat), https://www.edgepedia.org/fully-convolutional-network. Edgepedia Community License 1.0.", "credit_md": "\"[Fully convolutional network](https://www.edgepedia.org/fully-convolutional-network)\", Edgepedia (EdgeChat), [https://www.edgepedia.org/fully-convolutional-network](https://www.edgepedia.org/fully-convolutional-network). [Edgepedia Community License 1.0](https://www.edgepedia.org/edgepedia/license).", "credit_html": "\"Fully convolutional network\", Edgepedia (EdgeChat), https://www.edgepedia.org/fully-convolutional-network. Edgepedia Community License 1.0.", "speakable": "A fully convolutional network is a neural network architecture for image segmentation that replaces fully connected layers with convolutions, producing dense per-pixel predictions from inputs of any size." }