People who code: we want your input. Take the Survey

# Tag Info

3

CNNs like U-Net extract lower level features like edges on lower layers (i.e. the first convolutional layers) and higher level features on higher layers (i.e. convolutional layers closer to the final linear layers). This principle is losely inspired by how visual perception is implemented in the Visual Cortex among humans (and other animals). In a CNN the ...

2

These types of connections between layers are called skip connections, searching for this will give you a way to find more in-depth information. Broadly speaking, there are two advantages to using skip connections in this case: They allow gradients to more freely flow through the model, mitigating the issue of vanishing gradients They allow features from ...

2

It appears that the original images are 68x68 pixels and the model expects 256x256. You can use the Keras image processing API, in particular the smart_resize function to transform the images to expected number of pixels. Something like this: from tf.keras.preprocessing.image import smart_resize target_size = (256,256) image_resized = smart_resize(...

2

"Machine Learning for Digital Document Processing: from Layout Analysis to Metadata Extraction" by Esposito, Ferilli, Basile, and Mauro goes into detail about how to create a custom system for parsing digital documents, including pdfs. It proposes a generalized process to learn any structure within documents.

1

There is a library called facemesh for Python. It can detect face landmark points in Python. Connecting facemesh points to crop image to a desired polygon is trivial.

1

Whoa, are you trying to do clustering based on satellite images?! Just use the underlying long & lat coordinates to do this. It will be infinitely easier and so much more accurate too. Look at the example code below and try to adapt that to your specific case. import pandas as pd import numpy as np import matplotlib.pyplot as plt from sklearn.cluster ...

1

So as you already know, the convolutions extract some local information from the image, but for some decisions you need more global info. This is what pooling is supposed to do, it basically erases some locality information. In plain classification this is obviously necessary, because the question, if there is a cat in a picture, is a question about the ...

1

$\hspace{3cm}$ If we follow the definition of each arrow. Gray => Copy and Crop Every step in the expansive path consists of an upsampling of the feature map followed by a 2x2 convolution (“up-convolution”) that halves the number of feature channels, a concatenation with the correspondingly cropped feature map from the contracting path, and two 3x3 ...

1

I say that go with Object Detection. Because you already have annoted images. And segmentation applied to locate objects and boundaries (lines, curves, etc.) in images. . In Recent days EfficientDet Works very well in object detection. And for implementation look here. Best of luck.

1

Lets say my image is called frame and x, y, w and h are xmin, ymin, xmax and ymax You're confusing $w$ with $xmax$ and $h$ with $ymax$: Usually $w$ is the width of the crop whereas $xmax$ is the horizontal position of the end of the crop. Similarly $h$ is the height and $ymax$ is the vertical position of the end of the crop. Logically since $x$ is the (...

1

Diffgram is really great for this! I used it for a construction monitoring project. It's Open Source. From their site: Semantic Segmentation Tools: Auto Bordering: Automatically detects edges to create 100% coverage masks. Simple select the intersecting shape. Combo Shapes: Create shapes that are partially curves and partially straight lines. Points to Full ...

1

There can be a few reasons for this behavior as already pointed out in "Why is the validation accuracy fluctuating?": Size of train / validation stes: Fluctuations may become stronger the smaller your validation set is, especially during the early stage of training where predictions are closer to random predictions. Overfitting: Train loss seems ...

1

So the question asks about why different loss function lead to different error scores. So globally error is there to help us measure the level of discrimination between the output of the model and the actual output which we want to get. Different loss functions have different formulations of this and are thus depending on the task itself, more appropriate to ...

1

From the readme of your kaggle link; All images are provided in .tif format with 3 channels per image. For 101 cases, 3 sequences are available, i.e. pre-contrast, FLAIR, post-contrast (in this order of channels). For 9 cases, post-contrast sequence is missing and for 6 cases, pre-contrast sequence is missing. Missing sequences are replaced with FLAIR ...

1

As you can see in lines 286-296 in newmodels.py the model can use two different loss functions for the four different outputs. loss = {'pred1':lossfxn, 'pred2':lossfxn, 'pred3':lossfxn, 'final': losses.tversky_loss} loss_weights = {'pred1':1, 'pred2':1, 'pred3':1, 'final':1} model....

1

I think there is a bit of confusion here. The dice coefficient is defined for binary classification. Softmax is used for multiclass classification. Softmax and sigmoid are both interpreted as probabilities, the difference is in what these probabilities are. For binary classification they are basically equivalent, but for multiclass classification there is a ...

1

If your masks have only one channel and in the case of binary segmentation, you can easily compute a confusion matrix from one image thanks to: TP: ground truth and predicted pixel are of class 1 (object) TN: ground truth and predicted pixel are of class 0 (background) FP: ground truth pixel=0 while predicted pixel=1 FN: ground truth pixel=1 while predicted ...

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