Mask R-CNN | Practical Implementation

[@AmitChaudhary-qx5mc]

very easy and lucid explanation. Thank you mam

[@CodeWithAarohi]

Welcome 🙏

[@kaveeshamadhubhashana3094]

what are the data prprocesing features did u used

[@mariamnosheen4103]

Dear please tell if coco pretrain model did not support our custom datasets how can we develop our own custom datasets based pretrain model for mask rcnn. Do guide

[@accesspathology3654]

Thank you very much, your lecture saved me from darkness in working my thesis project. Would you mind to show us how to count the detected object as well? Thank you very much and hopefully everyone are protected from the pandemic. Om Swaha

[@CodeWithAarohi]

Will try to cover that part in my upcoming videos

[@outsiderladyrose]

Thank you very much for your straightforward explanation. I do have a question, can I use my 4 GB GTX 850 m for training? Thank you

[@CodeWithAarohi]

Yes you can!

[@accesspathology3654]

@@CodeWithAarohi Thank you very much for your reply. I do have 2 more queries, in case we need to count the number of detected object, how can we do that? and do you think utilize Google Colab will yield better and faster result? Thank you very much for your guidance, really appreciate it

[@sulmanahmed9348]

wow great i love your videos 😍

[@CodeWithAarohi]

Thankyou

[@samuelk9722]

thank you mam. could you make a video on Mask R-CNN model deployment using flask. or something like how to make an image segmentation API using flask or Django which can take image or videos as input

[@CodeWithAarohi]

Will surely make a video after finishing my pipelined videos.

[@omaraouat9727]

thank you , yours vidéos are so good

[@CodeWithAarohi]

Glad to know ...welcome

[@gayathrit.v8844]

what are the compatible versions of tensorflow and keras to run this code

[@CodeWithAarohi]

Try to work with this: kzbin.info/www/bejne/qmKwo62rdrqXpac This video have used the latest versions as compared to this one.

[@taniasultana5865]

Thank you so much for your informative video with the smart presentation. If the dataset has 2 types of data for example, MRI with Tumor= 100 and without tumor = 500. Then how can I prepare the data? Because without tumor data has no mask.

[@CodeWithAarohi]

You can train your data for 1 class ( brain with tumors ). No need for other class

[@taniasultana5865]

@@CodeWithAarohi Thank you for your kind information. I am trying this way.

[@nedjmahachemi990]

please give me solution of this error FutureWarning: Input image dtype is bool. Interpolation is not defined with bool data type. Please set order to 0 or explicitely cast input image to another data type. Starting from version 0.19 a ValueError will be raised instead of this warning. order = _validate_interpolation_order(image.dtype, order)

[@ragisimon4069]

Video is very informative and it is really helpful for my project. Ma'am ,Can you please explain how to create an interface for this mask rcnn model.

[@CodeWithAarohi]

Save your mask rcnn model and then you can create a web app using flask framework. BY creating a function in flask app, Load the saved model and make predictions

[@ragisimon4069]

@@CodeWithAarohi thank you ma'am...

[@ragisimon4069]

@@CodeWithAarohi Mam, can you please share code for interface.

[@sachinsrivastava968]

Well explained , thanks

[@CodeWithAarohi]

Welcome

[@sulmanahmed9348]

can i use Mask RCNN for road lane detection?

[@CodeWithAarohi]

Yes you can use

[@zahrahajalioghli8783]

dear Aarohi. Tnx for your helpful videos. if I have a dataset that has 10 different classes, each class in a different folder and each image just has one or two objects, can I use the VGG tool to annotate? which annotation tools you recommended?

[@CodeWithAarohi]

vgg annotator tool will give you a 1 json file for all the images( which will have annotation detail of each object in every Image).. You can put all the images of all classes in single folder and then use vgg annotator on it if you want to prepare the data and json file which I have used in this Mask RCNN example.

[@zahrahajalioghli8783]

@@CodeWithAarohi Thank you very much.i get it .

[@juliaestelleperrine2229]

Hi.. Can I know which python version, TensorFlow version, and Keras version you used in the Anaconda environment? I am having errors while running the code which tells that module 'Tensoflow' has no attribute log

[@CodeWithAarohi]

Tensorflow 1.14.1, keras 2.3.0

[@juliaestelleperrine2229]

@@CodeWithAarohi Thank you. It works fine now

[@makarandpawar5438]

@@juliaestelleperrine2229 what was your python version?

[@marwazaabi7320]

Thanks, it displays this error ? How i resolve it please: 'str' object has no attribute 'decode' (tensorflow 1.14.0, keras 2.3.0)

[@CodeWithAarohi]

Tell me the h5py version

[@marwazaabi7320]

@@CodeWithAarohi h5py version=3.1.0 and python version= 3.7.12

[@CodeWithAarohi]

@@marwazaabi7320 use h5py version 2.10.0

[@makarandpawar5438]

Hello Ma'am, I am getting "ImportError: cannot import name 'transpose_shape' from 'keras.utils.generic_utils' " while importing model from mrcnn. I'm using conda virtual environment with python 3.6.3, tensorflow 1.15 and keras 2.3.1 Please, help if you can!

[@joaofernandes6419]

Hey, there. Thank you for the explanations. I was following your code and I was doing an experiment with other images and I came across with this error while uploading the coco weights "Dimension 1 in both shapes must be equal, but are 12 and 324. Shapes are [1024,12] and [1024,324]. for 'Assign_684' (op: 'Assign') with input shapes: [1024,12], [1024,324]" How can I solve this? Thank you

[@suchitdreams]

Does mask rcnn works with tf 2 versions?

[@CodeWithAarohi]

Yes it works with tensorflow 2. But you have to modify this code for that

[@viplavanand5228]

Ma'am do we need to install tensoflow and cuda?

[@CodeWithAarohi]

Yes you need to install tensorflow version 1.

[@viplavanand5228]

@@CodeWithAarohi thank you ma'am

[@chidvilasyasannithi5689]

Ma'am this is so clear and I have followed everystep but I am getting an error saying : keyError: regions Please help me out I have sent you a mail regarding this and I have attached my code there, also descibed my error.

[@CodeWithAarohi]

keyError: regions means regions keyword is not present in your json file. Check your json file and use the keyword which is present in json file

[@shirsendudhar8944]

How can I print the confidence level in the output, not within the pictures?

[@CodeWithAarohi]

In mrcnn folder, there is one file with a name of visualize.py. That file is having all the code for displaying the mask, bounding boxes, class names and confidence score. You can do some modifications in that file for desired results.

[@shirsendudhar8944]

@@CodeWithAarohi Thanx

[@shahzebimtiaz1849]

I have successfully trained my model on a custom data set thanks to your previous video, I trained my model for 25 epochs so the logs folder has 25 files and the last one ends with maskrcnn00025h. so do I use that file from the logs directory for testing?

[@CodeWithAarohi]

Yes

[@ssagonline]

hi . i have some question regading this .. i have a datset of images on which i want the code to perform the feature selection based on a mathimatical formula ... can be help me regarding where i need to add the code for the feature selection... i would be a great help in my workd

[@CodeWithAarohi]

Then instead of using VGG-16, Use your Formula

[@ssagonline]

@@CodeWithAarohi can you please expand on that ... i am fairly new at working with this so it would help alot .. thank you

[@shirsendudhar8944]

In the test_model file, in-class CustomDataset, there is one class "object", with sub-classes laptop, mobile, and tab. What if we want to add another class "object_color" with sub_classes black, grey, and white. I think I have made necessary changes but getting stuck at self.add_image(). Any suggestions?

[@CodeWithAarohi]

Need to see your code. Then only can suggest you

[@shirsendudhar8944]

@@CodeWithAarohi How can I share it ?

[@CodeWithAarohi]

@@shirsendudhar8944 [e-mail]

[@janvitareddy5715]

Mam I am getting few errors while running the code,can you please help me in solving it?

[@CodeWithAarohi]

What error you are getting?

[@developerr4001]

Hello ma'am, I was having some issues with Mask RCNN implementation, hence I have sent u an email with the source code along with the dataset. Would you please be able to help?

[@CodeWithAarohi]

I will check your mail and revert you back there

[@developerr4001]

@@CodeWithAarohi Sure ma'am awaiting your reply :)

[@abhinitmahajan8462]

maam i got an error in the previous video , can u share your mail id so that i could send u the files . instead of classifying multiple objects im classifying just a single object

[@CodeWithAarohi]

[e-mail]

[@surflaweb]

I left my code to get the mAP, mAR and F1-Score github.com/matterport/Mask_RCNN/issues/2474 but for me it does not work.. Dear, when you trained your model the images had a size of fixed resolution? when I trained my model all my images were of different resolutions..can be this the problem? thanks in advance.

[@CodeWithAarohi]

I am not sure why its not working but you can share your code and sample of that data. I will try to correct it

[@yousseflejmi7678]

Can i send you my pictures by email to test it please ? It don't works with my laptop

[@CodeWithAarohi]

Its not just pictures. You need pictures with the json file. If you have both then send me

[@yousseflejmi7678]

@@CodeWithAarohi thank you for your feedback, I just shared with you the folder that contains the photos + json files (train and val), thank you for your help

[@PremKumar-zb8gt]

hello madam plase provise clarity videos

[@CodeWithAarohi]

Redo of this video: kzbin.info/www/bejne/qmKwo62rdrqXpac

[@nedjmahachemi990]

please give me solution of this error FutureWarning: Input image dtype is bool. Interpolation is not defined with bool data type. Please set order to 0 or explicitely cast input image to another data type. Starting from version 0.19 a ValueError will be raised instead of this warning. order = _validate_interpolation_order(image.dtype, order)