We release a light and portable version of Network Dissection in pyTorch at NetDissect-Lite. It is much faster than this first version and the code structure is cleaned up, without any complex shell commands. It takes about 30 min for a resnet18 model and 2 hours for a densenet161. If you have questions, please open issues at NetDissect-Lite.

Network Dissection is our method for quantifying interpretability of individual units in a deep CNN (i.e., our answer to question #1). It works by measuring the alignment between unit response and a set of concepts drawn from a broad and dense segmentation data set called Broden.

The BrodenHandle is a handle for datasets following the format of the Broad and Densely Labeled Dataset (Broden) dataset introduced in this paper. The original Broden dataset is not mandatory for usage of the provided handle. Any dataset following …

新数据集构建的基础是 Broadly Densely Labeled Dataset（Broden），这是一个包含不同视觉概念的混杂数据集。但是由于其设计初衷，Broden 并不适用于分割网络的训练。为此本文从 4 个方面做出优化，得到了 Broden+ 数据集： 去掉不同数据集的相似概念；

This is a pyTorch implementation of Unified Perceptual Parsing network on Broden+ dataset and ADE20K dataset. This work is published at ECCV'18 Unified Perceptual Parsing for Scene Understanding, to which Tete Xiao, Yingcheng Liu, and Bolei Zhou contribute equally.

Handles for Broden-like datasets. The original Broad and Densely Labeled Dataset (Broden) was introduced in [Bau2017] as a combination of several existing semantic segmentation and classification datasets on overlapping image sets. For more details on Broden and its encoding see BrodenHandle.

作者首先建立了一个完善的测试数据集，叫做Broden（Broadly and Densely Labeled Dataset），每张图片都在场景、物体、材质、纹理、颜色等层面有pixel-wise的标定。 接下来，将该数据集中的每一张图喂给需要分析的网络，拿到每个 feature map 上的响应结果，进一步分析该层feature map对应的语义关系，归纳结果。 整体流程如下图a所示。 a. 整体流程. b. Broden图片样例. 作者希望将每个 卷积核 单元（unit）与一些语义上的概念（concept）对应起来，从而使 …

2020年2月18日 · To identify a set of human-labeled visual concepts, the authors constructed the Broadly and Densely Labeled Dataset (Broden) dataset. The Broden dataset combines several densely labeled image...

2018年11月26日 · Load the segmentation dataset using the BrodenDataset class; use the transform_image argument to normalize images to be suitable for the model, and the size argument to truncate the dataset. Choose a directory in which to write the output, and call dissect(outdir, imodel, dataset) .

2023年7月19日 · 作者建立了一个完善的测试数据集，叫做Broden（Broadly and Densely Labeled Dataset），每张图片都在场景、物体、材质、纹理、颜色等层面有pixel-wise的标定。 Broden 数据集中的样本示例如下图所示。