11

There are 2 problems you might face. Your neural net (in this case convolutional neural net) cannot physically accept images of different resolutions. This is usually the case if one has fully-connected layers, however, if the network is fully-convolutional, then it should be able to accept images of any dimension. Fully-convolutional implies that it doesn'...


5

tl;dr What does that mean in the context of this paper? With "coarse segmentation" the author means a segmentation that doesn't have much detail. "Fine segmentation", on the other hand, refers to a segmentation with a high level of detail. But also more importantly [what does that mean in the context of] general computer vision? The most common use ...


3

Fully convolution networks A fully convolution network (FCN) is a neural network that only performs convolution (and subsampling or upsampling) operations. Equivalently, an FCN is a CNN without fully connected layers. Convolution neural networks The typical convolution neural network (CNN) is not fully convolutional because it often contains fully connected ...


3

The point is that in the expansive path you have two forms of information: the information from the contracting path, which includes all high-level features extracted from the original image. the information from the skip-connections, which copy a cropped version of the feature maps in the contracting path. Because, as we go forward through the expansive ...


3

I'd suggest you better understand edge detectors such as Robert or Sobel operators first to understand better how convolution operation on images extract features by constant value kernels. Would personally recommend Gonzales and Woods for this, as it gives a pure mathematical explanation to how and why these features are extracted. Essentially the ...


3

U-Net and U-Net inspired architectures have been quite popular in the medical image-related tasks ever since it was first introduced. There have been several improved versions of U-Net designed for specific tasks that followed. One such example is Attention U-Net, extremely popular for Pancreas Segmentation. Other examples of architectures that have ...


3

It depends on what is your ultimate goal. If your goal is to simply classify the object in the image, having more complex output won't help. Simpler output representation yields better result. If your goal is to detect the bounding box, output the bounding box. There is no need for a more complex output feature. If you use a segmentation method for bounding ...


2

See: Martin Thoma: A Survey of Semantic Segmentation, Section III Subsection A is about metrics and B is about datasets. Metrics include: accuracy, IoU, frequency weighted IoU, F-beta score, speed, ...


2

Deep learning based image segmentation is basically pixel-wise image classification. So instead of predicting a $C$ dimensional vector $\vec{x}$ where $C$ corresponds to the number of class labels, you predict a tensor $x\in\mathbb{R}^{h\times w \times C}$ of the same height and width dimensions as your input image, and with $C$ channels which correspond to ...


2

Disclaimer: This question is very broad, my answer is admittedly partial and is intended to just give an idea of what's out there and how to find out more. How do you "say" a network: "classify me these images" or "do semantic segmentation"? You're mixing two very different problems there. Although there are SO many variations of problems people are ...


2

CNN is used since it is effectively an optimized use case for dealing with image data. CNN effectively automatically extracts features from images. Other techniques are more likely to not take full advantage of the data. CNN is able to make full use of the data by also including information from adjacent pixels and downsample through layers. Here is a ...


2

Apart from the multitudes of traditional image segmentation techniques (Watershed, Clustering or Variational methods), newer Segmentation schemes using Deep Learning are actively being used, which provide better results and are better for real-time applications, owing to minimum computation overheads involved. The following blog provides a detailed review ...


2

If we seek proven working source code to plug into a GPLv2-licence compatible solution, we should at least consider autotrace. Its source code is open for review. It can be tested against the example images we have and, if it works fine, called by our GPLv2 software. We can even use the calling code in Inkscape's plug-in image tracing implementation as a ...


2

This is a really cool problem. You already have a working model here are a few different ways of going forward with the project. Grouping text based on locality. "no segmentation" Text region extraction in a document image based on the Delaunay tessellation Segmentation Multiscale Edge-Based Text Extraction from Complex Images Training a map of the ...


2

If you know it is symmetric, then you could do a couple things. Zero out a half. Don't bother learning both halves of the image. Just put a zero mask over the upper or lower half of the output matrix and just have the network regress the other half. Just don't make the network do more work than it needs to do. Learn both, but add symmetric loss In your ...


1

You can try doing image segmentation the traditional way, just using the image data. If you want to use the non-image data, then, you can introduce classification as another task for your network. It will provide some regularization to your model. But, this is one way you can still use non-image data whilst still working with image outputs.


1

Yes, $E$ is the cross-entropy function and a direct generalization of the binary case. For the binary case, probability to belong to the class $1$ is given by a sigmoid function $\sigma(x)$ of the output $x$, and the probability to belong to the class $0$ is $1 - \sigma(x)$. Therefore the binary crossentropy will give: $$ -\sum_i (l_i \log \sigma(x) + (1 - ...


1

Similar to other answers, I don't know Matlab that well but you could try the following steps to debug your problem. Make sure you can overfit to a single instance from your dataset, pull out a single image with a good amount of true positives in it. Duplicate that images B times (where B = Batch Size) and then try to train your network with only that small ...


1

I am afraid that the model will infer from the background information that it shouldn't use to predict the plant diseases, what makes the problem worse is that some plant diseases only exist in one dataset and not the other I am afraid that when I use the model on real-life conditions (say someone capturing an image with their phone) the model will be ...


1

In image segmentation the target is actually an image, with the same dimensions as the input, where each pixel has a label depending on which class it represents. It is not uncommon for such a dataset to have a "background" class that essentially consists of the pixels not belonging to any other class. If not you can always group together classes ...


1

It turned out that my intuition was not far off. The skeletonization is a good step. The Hough transform though is not a good way to create a graph of the roadmap. It seems that the Ramer–Douglas–Peucker algorithm can help out here. This algorithm first takes all the skeleton pixels as input, and sees this as a starting graph. The algorithm then proceeds to ...


1

This is a very hard problem, you have many overlapping points with objects which aren't completely round. I'm not very knowledgeable on CV but I suspect you will find it very challenging. I would probably say a handcrafted detection algorithm would probably be easier, something like an edge detector which fit circles to arcs and labeled the points. But it's ...


1

I found a method to do it in the paper Cross-Modality Personalization for Retrieval (2020, accessed: 20-Feb-2020). Representation. For images, we extract Inception-v4 CNN features [36]. We then mask the image convolution feature with the BubbleView saliency map, by resizing the saliency map to the convolution feature size and multiplying them together. ...


1

You can find leaderboards as well as code at this address. For now, HRNetV2 leads the game. The U-Net architecture is part of a broad family of network architectures that aggregate multi-scale features to extract finer details useful for semantic segmentation. Examples are Feature Pyramidal Networks (FPN), Hourglass, Encoder-Decoder, MatrixNet, etc...


1

It sounds like you are interested in the ideas of intrinsic motivation and attention in the context of machine learning. These are big topics, and the subject of much active research. Intrinsic motivation says that the key to identifying interesting patterns and skills that are worth learning is to give the agent some intrinsic reason to learn to do new ...


1

If your interest is positional information, encode it! This could include learning an embedding for each position and leveraging that in your model. You could also use an approach to hard-encode rather than learn it (kinda like adding sinusoids in the transformer paper Attention is All You Need an example of a paper that encodes the 2D positional info: ...


1

The paper referenced by Martin Thoma is the go-to for semantic segmentation. However I will also like to add the Panoptic Segmentation metric as an aggregated method to measure both the detection task and segmentation task of the model. It is a very well-known and widely used metric since it is the standard metric for COCO dataset (segmentation) This is the ...


1

Try resizing the image to the input dimensions of your neural network architecture(keeping it fixed to something like 128*128 in a standard 2D U-net architecture) using nearest neighbor interpolation technique. This is because if you resize your image using any other interpolation, it may result in tampering with the ground truth labels. This is particularly ...


1

As you want to perform image segmentation, you can use U-Net, which does not have fully connected layers, but it is a fully convolutional network, which makes it able to handle inputs of any dimension. You should read the linked papers for more info.


1

You could also have a look at the paper Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition (2015), where the SPP-net is proposed. SSP-net is based on the use of a "spatial pyramid pooling", which eliminates the requirement of having fixed-size inputs. In the abstract, the authors write Existing deep convolutional neural ...


Only top voted, non community-wiki answers of a minimum length are eligible