I am working on a classification task using transfer learning. I am using ResNet50 and weights from ImageNet.
My_model = (ResNet50( include_top=False, weights='imagenet', input_tensor=None,
input_shape=(img_height, img_width, 3),pooling=None))
I didn't rescale my input images between 0-255 but my result is quite good (acc: 93.25%). So my question is do I need to rescale images between 0-255? Do you think my result is wrong without rescaling between 0-255?
Thank you.
No basically your result is not wrong. to give a clue on that, we standardize the pixels values to a range between (0 and 1) just to avoid resulting big values during the calculus in the forward propagation z = w*x + b and then the backward propagation.
Why we do that ?
I develop, the optimization algorithm is definitely dependent on the result of the backward prop, so when we start updating our optimization algo with big values of weights/bias, then we need then a lot of epochs to reach the global minimum.
Related
I build a LSTM to forecast Precipitation, but it doesn't work well.
My code is very simple and data is very short only contains 720 points.
i use MinMaxScale to scale the data.
this is my code, seq_len = 12
model = Sequential([
layers.LSTM(2, input_shape=(SEQ_LEN, 1),
layers.Dense(1)])
my data is like this
and the output compares with true value like this
I use adam and mae loss function, epoch=10
is it underfitting? or is this simple net can't do this work?
r2_score is no more than 0.55
please tell me how to adjust it. thanks
there are so many options;
first of all it would be better to define the optimized window size by changing the periods of the sequences
The second option would be changing the batch-size of the dataset
Change optimizer into SGD cause of few datapoints and before training model define the best values for learning rate by setting Learning Rate Schedule callback
Try another model architecture with convolution layers and etc
Sometimes it would be a trick to help model performance by setting lambda layer after the last layer to scale up values cause of lstm default activation function is tanh.
I'm studying on a deep learning(supervised-learning) to estimate depth images from monocular images.
And the dataset currently uses KITTI data. RGB images (input image) are used KITTI Raw data, and data from the following link is used for ground-truth.
In the process of learning a model by designing a simple encoder-decoder network, the result is not so good, so various attempts are being made.
While searching for various methods, I found that groundtruth only learns valid areas by masking because there are many invalid areas, i.e., values that cannot be used, as shown in the image below.
So, I learned through masking, but I am curious about why this result keeps coming out.
and this is my training part of code.
How can i fix this problem.
for epoch in range(num_epoch):
model.train() ### train ###
for batch_idx, samples in enumerate(tqdm(train_loader)):
x_train = samples['RGB'].to(device)
y_train = samples['groundtruth'].to(device)
pred_depth = model.forward(x_train)
valid_mask = y_train != 0 #### Here is masking
valid_gt_depth = y_train[valid_mask]
valid_pred_depth = pred_depth[valid_mask]
loss = loss_RMSE(valid_pred_depth, valid_gt_depth)
As far as I can understand, you are trying to estimate depth from an RGB image as input. This is an ill-posed problem since the same input image can project to multiple plausible depth values. You would need to integrate certain techniques to estimate accurate depth from RGB images instead of simply taking an L1 or L2 loss between an RGB image and its corresponding depth image.
I would suggest you to go through some papers in estimating depth from single images such as: Depth Map Prediction from a Single Image using a Multi-Scale Deep Network where they use a network to first estimate the global structure of the given image and then use a second network that refines the local scene information. Instead of taking a simple RMSE loss, as you did, they use a scale-invariant error function in which the relationship between points is measured.
I'm a beginner in CNN DeepLearning, I know the basic concept that we use some filters to generate a set of feature maps from an image, we activate it using non-linear method like 'relu' before we downsample it. We keep doing this until the image becomes very small. Then we flatten it and use a fully connected network to calculate its category. And we use the back-propergation technique to calculate all parameters in the map. One thing I don't understand is that when we do Conv2D we create many filters(channels) from an image. Like in the sample code:
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(150, 150, 3)))
I understand this is to generate as many features as possible. But how these filters are trained to detect different features from one image? If all of them are initialized with the same value (like 0) then they should end up with detecting the same feature, right? Are we giving them random values during initialization so that they can find their local minimum loss using gradient descent?
If you initialize all filters with the same value, then you are right, they will learn the same thing. That's why we never initialize with same value. We initialize each kernel with random values (usually 0 mean and some small variance).
There are many methods to find out a good initialization for your network. One of the most famous and used ones is Xavier initialization.
Adding to what being discussed, the weights in the CONV layer also learns the same way weights learn in FC layer, through backpropagation, using some optimization algorithm (GD, Adam, RMSprop etc). Ending up in local optimum is very unlikely in big networks as a point being local optimum for all the weights is very unlikely as no of weights increases. If weights are initialized with zeros, the gradients become the same for the update and hidden units become the same in a layer. Hence they learn the same features. Hence we use random initialization with mean 0 and variance inversely proportional to the number of units in the previous layer. (eg Xavier)
I want to fit a curve to data obtained from an FFT. While working on this, I remembered that an FFT gives binned data, and therefore I wondered if I should treat this differently with curve-fitting.
If the bins are narrow compared to the structure, I think it should not be necessary to treat the data differently, but for me that is not the case.
I expect the right way to fit binned data is by minimizing not the difference between values of the bin and fit, but between bin area and the area beneath the fitted curve, for each bin, such that the energy in each bin matches the energy in the range of the bin as signified by the curve.
So my question is: am I thinking correctly about this? If not, how should I go about it?
Also, when looking around for information about this subject, I encountered the "Maximum log likelihood" for example, but did not find enough information about it to understand if and how it applied to my situation.
PS: I have no clue if this is the right site for this question, please let me know if there is a better place.
For an unwindowed FFT, the correct interpolation between bins is by using a Sinc (sin(x)/x) or periodic Sinc (Dirichlet) interpolation kernel. For an FFT of samples of a band-limited signal, thus will reconstruct the continuous spectrum.
A very simple and effective way of interpolating the spectrum (from an FFT) is to use zero-padding. It works both with and without windowing prior to the FFT.
Take your input vector of length N and extend it to length M*N, where M is an integer
Set all values beyond the original N values to zeros
Perform an FFT of length (N*M)
Calculate the magnitude of the ouput bins
What you get is the interpolated spectrum.
Best regards,
Jens
This can be done by using maximum log likelihood estimation. This is a method that finds the set of parameters that is most likely to have yielded the measured data - the technique originates in statistics.
I have finally found an understandable source for how to apply this to binned data. Sadly I cannot enter formulas here, so I refer to that source for a full explanation: slide 4 of this slide show.
EDIT:
For noisier signals this method did not seem to work very well. A method that was a bit more robust is a least squares fit, where the difference between the area is minimized, as suggested in the question.
I have not found any literature to defend this method, but it is similar to what happens in the maximum log likelihood estimation, and yields very similar results for noiseless test cases.
I was wondering which is the best machine learning technique to approximate a function that takes a 32-bit number and returns another 32-bit number, from a set of observations.
Thanks!
Multilayer perceptron neural networks would be worth taking a look at. Though you'll need to process the inputs to a floating point number between 0 and 1, and then map the outputs back to the original range.
There are several possible solutions to your problem:
1.) Fitting a linear hypothesis with least-squares method
In that case, you are approximating a hypothesis y = ax + b with the least squares method. This one is really easy to implement, but sometimes, a linear model is not good enough to fit your data. But - I would give this one a try first.
Good thing is that there is a closed form, so you can directly calculate parameters a and b from your data.
See Least Squares
2.) Fitting a non-linear model
Once seen that your linear model does not describe your function very well, you can try to fit higher polynomial models to your data.
Your hypothesis then might look like
y = ax² + bx + c
y = ax³ + bx² + cx + d
etc.
You can also use least squares method to fit your data, and techniques from the gradient descent types (simmulated annealing, ...). See also this thread: Fitting polynomials to data
Or, as in the other answer, try fitting a Neural Network - the good thing is that it will automatically learn the hypothesis, but it is not so easy to explain what the relation between input and output is. But in the end, a neural network is also a linear combination of nonlinear functions (like sigmoid or tanh functions).