yolov3 upsampleLayer problem.

I wanna follow "Object Detection Using YOLO v3 Deep Learning example.

But there is not funciton defining "upsampleLayer".

I tried to use "transposedConv2dLayer" with the following specs in the "addSecondDetectionHead.m"

>>>

weights = zeros(2,2,numFilters,numFilters);

bias = zeros(1,1,numFilters);

for i= 1:numFilters

weights(:,:,i,i) = 1;

end

transposedConv2dLayer([2,2],numFilters,'Stride',[2,2],...

'WeightLearnRateFactor',0,'BiasLearnRateFactor',0, 'Name', 'ups_2d', 'Weights', weights, 'Bias',bias );

<<<

But I get the following error:

***

Error using dlfeval (line 43)

Layer 'ups_2d': Invalid input data. Number of channels to convolve (18, specified by the size

of weights dimension number 4) must be equal to the size of the 'C' dimension of the input data

(256).

Any upsamplelayer function working ?

I also want to check each layer input and output size, is there any practical solution ?

[Joon]

I have same problem. I would like to ask to MATHWORKS to upload 'upsampleLayer' function.

[Aasim Khurshid]

This function is part of the example of object detection. You should go to the example in MATLAB (click the "View MATLAB Command" link in the box in the upper-right corner of the example or open it in MATLAB Online by clicking the button in that same box) and you will find that helper in the example's directory. Or you may click on the error, where it says, it is used in object detection example, Click there, Copy this file to your directory, and you are good to go.

Do the same for other missing files as well.

Indicated here for another file:

https://www.mathworks.com/matlabcentral/answers/608676-validateinputdata-in-yolo-v3-missing

Answer 1

[SERDAR KIZILKAYA]

Function is built in Matlab/examples, also given below:

% Upsample by replicating neighbouring pixel values.

% Copyright 2019 The MathWorks, Inc.

classdef upsampleLayer < nnet.layer.Layer

properties

% factor to upsample the input.

UpSampleFactor

end

methods

function layer = upsampleLayer(factor, name)

% Set layer name.

layer.Name = name;

% Set layer description.

layer.Description = "upSamplingLayer with factor " + factor;

% Stride by which layer is upsampled.

layer.UpSampleFactor = factor;

end

function Z = predict(layer, X)

% Z = predict(layer, X) forwards the input data X through the

% layer and outputs the result Z.

Z = repelem(X,layer.UpSampleFactor,layer.UpSampleFactor);

end

end

end

[Joon]

Thanks for uploading.

For 'upsampleLayer' it works fine. However, I got the error message when I run the follwing code:

% Convert layer graph to dlnetwork.

net = dlnetwork(lgraph);

The error message says that 'conv2Detection1' is the output which is not used and this must be connected to the input of other layer.

Could you help me to run the example code?

Added: I also encountered the error that 'generateTargets' function is not found (for detailed error message, see below). I am wondering whether all these errors including previous errors are because I am using R2019b version? It is possible to install 2020 version if necessary.

'generateTargets' Error: matlab_yolov3_example>modelGradients (line 254)

[boxTarget, objectnessTarget, classTarget, objectMaskTarget,

boxErrorScale] = generateTargets(gatheredPredictions, YTrain,

inputImageSize, anchors, mask, penaltyThreshold);

[AODI]

I encountered all these errors as you.(R2019b)

Then I installed 2020a version and all the problems disappeared!

So exciting~

[Joon]

Thanks! I just gave up and have been using Yolov2 instead. But it is working after I installed 2020a as you suggested. Cool!

[Joon-Hyuk Lee]

I got another problem during the training. The value of bboxA (which is the predicted box, not truth box) seems to have infinite(?) value. Anyone had same problem before?

Error messages are here:=====================================

Error using bboxOverlapRatio

The value of 'bboxA' is invalid. Expected input to be finite.

Error in bboxOverlapRatio>validateAndParseInputs (line 195)

parser.parse(bboxA,bboxB,varargin{:});

Error in bboxOverlapRatio>iParseInputs (line 94)

[bboxA, bboxB, ratioType] = validateAndParseInputs(bboxA, bboxB, varargin{:});

Error in bboxOverlapRatio (line 55)

[bboxA, bboxB, ratioType, isUsingCodeGeneration] = iParseInputs(bboxA,bboxB,varargin{:});

Error in generateTargets>getMaxIOUPredictedWithGroundTruth (line 138)

overlap = bboxOverlapRatio(predb,truthBatch);

Error in generateTargets (line 45)

iou = getMaxIOUPredictedWithGroundTruth(bx,by,bw,bh,groundTruth);

Error in yolov3>modelGradients (line 354)

[boxTarget, objectnessTarget, classTarget, objectMaskTarget, boxErrorScale] = generateTargets(gatheredPredictions, YTrain, inputImageSize, anchors, mask, penaltyThreshold);

Error in deep.internal.dlfeval (line 18)

[varargout{1:nout}] = fun(x{:});

Error in dlfeval (line 41)

[varargout{1:nout}] = deep.internal.dlfeval(fun,varargin{:});

==================================================