The MATLAB function xcepption apprears to return a xception-41 netowrk archtecture , not the xception-71 network architecture

4 次查看（过去 30 天）
Chen Xinbing
Chen Xinbing 2023-9-27
回答： Jayanti 2024-10-17
I have tested the MATLAB function xception using the following code:
analyzeNetwork(layerGraph(xception()));
I find that the returned network by xception() is a network architecure of xception-41(from Chollet, F., 2017. "Xception: Deep Learning with Depthwise Separable Convolutions." arXiv preprint, pp.1610-02357.), not the network architecure of xception-71(from Chen, L., Y. Zhu, G. Papandreou, F. Schroff, and H. Adam. "Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation." Computer Vision — ECCV 2018, 833-851. Munic, Germany: ECCV, 2018.). But in the help document for xception, it is described as "Xception is a convolutional neural network that is 71 layers deep". My question is if does the document appear to make a mistake?

请先登录，再回答此问题。

采纳的回答

Jayanti
Jayanti 2024-10-17
Hi Chen,
The architecture returned by “xception()” is working as intended. It is same as the model given in paper “Chollet, François. “Xception: Deep Learning with Depthwise Separable Convolutions.” Preprint, submitted in 2016”(https://arxiv.org/abs/1610.02357).
The “Xception” network uses depth wise separable convolutions, which are implemented in MATLAB as pairs of regular convolutional layers. Under MATLAB's definition of depth, counting each convolution layer separately, the network reaches a depth of 71. However, if you count each pair as a single depth wise separable convolution, you get 41 layers, which matches the architecture of given paper.
In summary, this confusion arises due to different counting nomenclature.

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by