ttmronald / aerialcrackdetection Goto Github PK

I want to run my own dataset on your code. I make same structure as you did.
But I have some questions.
my dataset image sizes should be the same with your dataset image sizes? or it is doesn't matter?

Thank you!

Hi @TTMRonald ,

Thanks for sharing this awesome work, I had some doubts about setting the parameters in config.py.
img_channel_mean, as per the name, should I be putting here channel wise means??
img_scaling_factor, How to find this?
classifier_regr_std , how to find this?
std_scaling, how to find this?
rpn_stride, I know this is related to the network architecture we select but how to find it??

Your works is very interesting. so I would like to read and cite your paper, "AerialCrackDataset: Towards Object Detection with Dataset". but I couldn't find the paper.
Could you send or upload the paper? thank you.
email: [email protected]

File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1864, in _create_c_op c_op = c_api.TF_FinishOperation(op_desc) tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimension 0 in both shapes must be equal, but are 3 and 512. Shapes are [3,3,2048,1024] and [512,1024,3,3]. for 'Assign' (op: 'Assign') with input shapes: [3,3,2048,1024], [512,1024,3,3].

I'm getting this error when I run the test script with many of the network types. I'm getting it with your AerialCrackDataset and with my own images. I get it when using GoogleNet, ZF, or VGG16 as the Network parameter.

Also, when using ResNet101 on your images I get output like this indicating that no cracks were detected:
000001.jpg Elapsed time = 6.913724184036255 []

I do have it working for ResNet50.

Hi. I followed the instructions to train the network and it worked without an issue. I used the resnet50 to train the network and then try to execute the ./test.py file, however I am running into the same issue repeatedly:

Traceback (most recent call last):
File "test.py", line 50, in
config_output_filename = C.config_filename
NameError: name 'C' is not defined

I've gone through the train.py file and tried to bring over the same type of declaration used in that file:

C = config.Config()

But even this doesn't seem to be helping me. Does anyone know how I can move past this or did I do something wrong? I will be continue trying to fix this issue over the next couple days and any help would be GREATLY appreciated. I'll also be sure to update my progress, but I've been stuck on this issue for a while now and it is frustrating. Thanks in advance for your help.

Hello TTMRonald,
Your work is really interesting. I saw your results in the Caffe version repo and your results are actually promising. However I've got 2 questions for you.

1. I don't have a fast internet connection, so I still haven't tried this or the caffe version. But I digress. what I'm about to ask is if you have demo pictures for this Keras version, like the demo pictures you posted for the caffe version.

2. I'm not yet an expert in RCNNs and therefore I don't know yet there difference with each other(Simple RCNN, FastRCNN, FasterRCNN) but I see in this repo (Keras version) that you named one of your folder as FastRCNN instead of FasterRCNN. Is this just a mistake, or did you really use jsut FastRCNN and not FasterRCNN?

Thank you and regards,
Rusty

Google Drive ask for permission