I'm using one-vs-all to do a 21-class svm categorization.
I want the label -1 to mean "not in this class" and the label 1 to mean "indeed in this class" for each of the 21 kernels.
I've generated my pre-computed kernels and my test vectors using this standard.
Using easy.py everything went well for 20 of the classes, but for one of them the labels were switched so that all the inputs that should have been labelled with 1 for being in the class were instead labelled -1 and vice-versa.
The difference in that class was that the first vector in the pre-computed kernel was labelled 1, while in all the other kernels the first vector was labelled -1. This suggests that LibSVM relabels all of my vectors.
Is there a way to prevent this or a simple way to work around it?
You already discovered that libsvm uses the label -1 for whatever label it encounters first.
The reason is, that it allows arbitrary labels and changes them to -1 and +1 according to the order in which they appear in the label vector.
So you can either check this directly or you look at the model returned by libsvm.
It contains an entry called Label which is a vector containing the order in which libsvm encountered the labels. You can also use this information to switch the sign of your scores.
If during training libsvm encounters label A first, then during prediction
libsvm will use positive values for assigning object the label A and negative values for another label.
So if you use label 1 for positive class and 0 for negative, then to obtain right output values you should do the following trick (Matlab).
%test_data.y contains 0-s and 1-s
[labels,~,values] = svmpredict(test_data.y, test_data.X, model, ' ');
if (model.Label(1) == 0) % we check which label was encountered by libsvm first
values = -values;
end
Related
I am trying to use DeepLabV3 for image segmentation and object detection/classification on Coral.
I was able to sucessfully run the semantic_segmentation.py example using DeepLabV3 on the coral, but that only shows an image with an object segmented.
I see that it assigns labels to colors - how do i associate the labels.txt file that I made based off of the label info of the model to these colors? (how do i know which color corresponds to which label).
When I try to run the
engine = DetectionEngine(args.model)
using the deeplab model, I get the error
ValueError: Dectection model should have 4 output tensors!This model
has 1.
I guess this way is the wrong approach?
Thanks!
I believe you have reached out to us regarding the same query. I just wanted to paste the answer here for others to reference:
"The detection model usually have 4 output tensors to specifies the locations, classes, scores, and number and detections. You can read more about it here. In contrary, the segmentation model only have a single output tensor, so if you treat it the same way, you'll most likely segfault trying to access the wrong memory region. If you want to do all three tasks on the same image, my suggestion is to create 3 different engines and feed the image into each. The only problem with this is that each time you switch the model, there will likely be data transfer bottleneck for the model to get loaded onto the TPU. We have here an example on how you can run 2 models on a single TPU, you should be able to modify it to take 3 models."
On the last note, I just saw that you added:
how do i associate the labels.txt file that I made based off of the label info of the model to these colors
I just don't think this is something you can do for segmentation model but maybe I'm just confused on your query?
Take object detection model for example, there are 4 output tensors, the second tensor gives you an array of id associates with a certain class that you can map to a a label file. Segmentaion models only give the pixel surrounding an objects.
[EDIT]
Apology, looks like I'm the one confused on segmentation models.
Quote form my college :)
"You are interested to know the name of the label, you can find the corresponding integer to that label from result array in Semantic_segmentation.py. Where result is classification data of each pixel.
For example;
if you print result array in the with bird.jpg as input you would find few pixel's value as 3 which is corresponding 4th label in pascal_voc_segmentation_labels.txt (as indexing starts at 0 )."
I'm training Deeplab v3 by making custom data set in three class, including background
Then, My class is background, panda, bottle and there are 1949 pictures.
and I'm using a moblienetv2 model
and segmentation_dataset.py has been modified as follow.
_MYDATA_INFORMATION = DatasetDescriptor(
splits_to_sizes={
'train': 975, # num of samples in images/training
'trainval': 1949,
'val': 974, # num of samples in images/validation
},
num_classes=3,
ignore_label=0,
)
train.py has been modified as follow.
flags.DEFINE_boolean('initialize_last_layer', False,
'Initialize the last layer.')
flags.DEFINE_boolean('last_layers_contain_logits_only', True,
'Only consider logits as last layers or not.')
train_utils.py has not been modified.
not_ignore_mask = tf.to_float(tf.not_equal(scaled_labels, ignore_label)) * loss_weight
I get some results, but not the perfect ones.
For example, the mask colors of panda and bottles are same or not distinct
The result that I want is panda of red and bottle of green
So, I judged that there was a problem with the weight.
Based on the other people's questions, train_utils.py was configured as follows
irgore_weight = 0
label0_weight =1
label1_weight = 10
label2_weight = 15
not_ignore_mask =
tf.to_float(tf.equal(scaled_labels, 0)) * label0_weight +
tf.to_float(tf.equal(scaled_labels, 1)) * label1_weight +
tf.to_float(tf.equal(scaled_labels, 2)) * label2_weight +
tf.to_float(tf.equal(scaled_labels, ignore_label)) * irgore_weight
tf.losses.softmax_cross_entropy(
one_hot_labels,
tf.reshape(logits, shape=[-1, num_classes]),
weights=not_ignore_mask,
scope=loss_scope)
I have a question here.
What are the criteria for the weight?
My data set consists of the following.
enter image description here
It's automatically generating, so I don't know exactly which one is more, but it's a similar amount.
And another thing, I'm using Pascal's color map type.
This is the first black background and the second red third green.
I want to designate pandas as red and bottles as green exactly. What should I do?
I think you might have mixed up your label definition. Maybe I can help you with that. Please check again your segmentation_dataset.py. Here, you define "0" as the ignored label. This means that all pixels which are labeled as "0" are excluded from the training process (more specifically, excluded in the calculation of the loss function and so have no influence in the updating of the weights). In the light of this situation it is crucial to not "ignore" the background class as it is also a class you want to predict correctly. In train_utils.py you assign a weightening factor to the ignored class which would have no effect- --> Make sure that you don't mix up your three training classes [background, panada, bottle] with the "ignored" tag.
In your case num_classes=3 should be correct as it specifies the number of labels to predict (the model automatically assumes these labels are 0, 1 and 2. If you want to ignore certain labels you have to annotate them with a fourth label class (just choose a number >2 for that) and then assign this label to ignored_label. If you don't have pixels to be ignored still set ignored_label=255 and it will not influence your training;)
I'm experimenting with tensorflow for speech recognition.
I have inputs as waveforms and words as output.
The waveform would look like this
[0,0,0,-2,3,-4,-1,7,0,0,0...0,0,0,20,-11,4,0,0,1,...]
The words would be an array of numbers while each number represents a word:
[12,4,2,3]
After training I also want to find out the correlation between input and output for each output label.
For example I want to know which input neurons | samples are responsible for the first label (here 12).
[0,0.01,0.10,0.99,0.77,0.89,0.99,0.79,0.22,0.11,0...0,0,0,0,0,0,0,0,0,...]
The original values of the input would be replaced with the correlation while 0 means no correlation and 1 means total correlation.
The goal is to get the position when a word starts.
Is there a function in tensorflow to get this correlation?
Question
I have a sequence of data (X) that I want to translate into another sequence of data (Y) as well as report what part of (X) contributed to (Y).
Answer
This is a well known problem and Tensorflow.org actually has a fantastic example neural machine translation with attention
The example code show how to translate X (Spanish) into Y (English) and report what part of X contributes to the decision of each part of Y (attention)
The exact same principle and code can be used to translate X (wave data) into Y (words) and report what part of the wave data contributes to each word via the attention readout.
The attention layer in the example is called attention_layer.
I currently want to use Tensorflows Object Detection API for my custom problem.
I already created the dataset, but its pretty unbalanced.
The Dataset has 3 classes and my main problem is, that one class has about 16k samples and another class has only about 2.5k samples.
So I think I have to balance the dataset. Someone told me, that there is something called sample/class weights(Not sure if this is 100% correct), which balance the samples for training, so that the biggest class has a smaller impact on training then the smallest class.
I'm not able to find this method for balancing. Can someone pleas give me a hint where to start?
You can do normal cross entropy, giving you a ? x 1 tensor, X of losses
If you want class number N to count T times more, you can do
X = X * tf.reduce_sum(tf.multiply(one_hot_label, class_weight), axis = 1)
tf.multiply
scales the label by whatever weight you want,
tf.reduce_sum
converts the label vector a to a scalar, so you end up with a ? x 1 tensor filled with the class weightings. Then you simply multiply the tensor of losses with the tensor of weightings to achieve desired results.
Since one class is 6.4 times more common than the other, I would apply the weightings 1 and 6.4 to the more common and less common class respectively. This will mean that every time the less common class occurs, it has 6.4 times the affect of the more common class, so it's like it saw the same number of samples from each.
You might want to modify it so that the weighting add up to the number of classes. This matches the default case is all of the weightings are 1. In that case we have 1 /7.4 and 6.4/7.4
nce_loss() asks for a static int value for num_true. That works well for problems where we have the same amount of labels per training example and we know it in advance.
When labels have a variable shape [None], and being batched and/or bucketed by bucket size with .padded_batch() + .group_by_window() it is necessary to provide a variable size num_true in order to accustom for all training examples. This is currently unsupported to my knowledge (correct me if I'm wrong).
In other words suppose we have either a dataset of images with an arbitrary amount of labels per each image (dog, cat, duck, etc.) or a text dataset with numerous multiple classes per sentence (class_1, class_2, ..., class_n). Classes are NOT mutually exclusive, and can vary in size between examples.
But as the amount of possible labels can be huge 10k-100k is there a way to do a sampling loss to improve performance (in comparison with a sigmoid_cross_entropy)?
Is there a proper way to do this or any other workarounds?
nce_loss = tf.nn.nce_loss(
weights=nce_weights,
biases=nce_biases,
labels=labels,
inputs=inputs,
num_sampled=num_sampled,
# Something like this:
# `num_true=(tf.shape(labels)[-1])` instead of `num_true=const_int`
# , would be preferable here
num_classes=self.num_classes)
I see two issues:
1) Work with NCE with different numbers of true values;
2) Classes that are NOT mutually exclusive.
To the first issue, as #michal said, there is an expectative of including this functionality in the future. I have tried almost the same thing: to use labels with shape=(None, None), i.e., true_values dimension None. The sampled_values parameter has the same problem: true_values number must be a fixed integer number. The recomended work around is to use a class (0 is the best one) representing <PAD> and complete the number of true_values. In my case, 0 is an special token that represents <PAD>. Part of code is here:
assert len(labels) <= (window_size * 2)
zeros = ((window_size * 2) - len(labels)) * [0]
labels = labels + zeros
labels.sort()
I sorted the label because considering another recommendation:
Note: By default this uses a log-uniform (Zipfian) distribution for
sampling, so your labels must be sorted in order of decreasing
frequency to achieve good results.
In my case, the special tokens and more frequent words have lower indexes, otherwise, less frequent words have higher indexes. I included all label classes associated to the input at same time and completed with zero till the true_values number. Of course, you must ignore the 0 class at the end.