Faster R-CNN on Custom Dataset | Custom Object Detector

[@poojithsrisai2576]

You are the saviour for many students madam. thank you

[@CodeWithAarohi]

It's my pleasure 🙂

[@infinityX12]

I am using roboflow for my dataset labelling, in what format should i download the dataset?

[@iamjaseer791]

how can we convert our yolo dataset to faster rcnn model. or how can we create faster rcnn data using custom images we have

[@ahmedabbas3981]

Upload dataset to roboflow and export it in faster rcnn format

[@vemulasuman6995]

Excellent ...explained in very understaandable way

[@CodeWithAarohi]

Glad you liked it

[@piyushdubey_10]

You are awesome mam . I trained models after your videos. thank you

[@CodeWithAarohi]

My pleasure 😊

[@moumabenaf940]

@piyushdubey_10 Can you show me how you trained the model please

[@gowthamkumar3296]

Great video ! I have gone through the code. But I didn't find the training step .... can you please explain where the training step is ??

[@rajmrittik1012]

your videos are just so convenient

[@CodeWithAarohi]

Thank you!

[@ayushisaxena748]

Thankyou mam you are the best 🙏🙏🙏🙏I have started just in this field with your videos only I am gaining confident...Thankyou Mam ❤️

[@CodeWithAarohi]

Most welcome 😊

[@dswithgagan2110]

thanks Aarohi for giving these tutorials, these are really very helpful

[@CodeWithAarohi]

Welcome

[@TechLearner81]

In the start till RPN everything goes smoothly and I'm happyt to finally such great videos which explains the code alongwith theoretical concepts. But at the all goes in vain there's no explanation what you have teached can be used practically to train the model.

[@shalinig1253]

Got infinite loop while run this line of code(X,Y,image_data,debug_img, debug_num_pos=next(data_gen_train)) in the training_fasterrcnn.ipynb file?

[@CodeWithAarohi]

share your code on my email id

[@petermungai5508]

Same to me. Did you find solution?

[@petermungai5508]

Did you find any solution

[@petermungai5508]

Hi Shalini . Did you find solution to infinite loop?

[@EbookMemories]

Mam, to test our modelling how to test our modeling is successful or not, we must testing the new images that haven't been annotated, how to make it like that?

[@EbookMemories]

Mam, how to provide a bounding box in the detected image and provide accuracy for each image given the bounding box.

[@AagneyWest]

Try roboflow

[@petermungai3557]

When I run this code. Jupyter notebook grows in size and crush. Is the a way I can save train data generator somewhere else X, Y, image_data, debug_img, debug_num_pos = next(data_gen_train) and latter load them without making the jupyter notebook to grow in size

[@devavratpro7061]

Hi Arohi, thanks for this great video. How can I prepare my own training dataset? Which tool to use for it? Can we use VGG annotator (with which bounding box format/shape)? Kindly share a chain of training dataset preparation starting from annotation to achieving the required Faster RCNN format?

[@habil183]

have you know how to do it?

[@brunojerkovic8666]

LOVEE THIS VIDEO. Finally, GREAT explanation on how to do this :) Just one thing, Does this code contain code for both training and testing the model or only training? Also, if training/testing is not covered in the video, is there another video explaining how to do training/testing on a custom dataset? I need to train the Faster R-CNN model, do you recommend any of your other videos for that? I have also watched all your videos about Faster R-CNN, I am mostly concerned about how to train and test it.

[@CodeWithAarohi]

Glad you liked my video. And this is the only 1 video for custom dataset. I hadn't prepared the further part yet.

[@rabomeister]

@@CodeWithAarohi You did not answer the question. He basicly asked that whether you explicitly explained how to seperate train and test and furthermore, however you just explained the title of the video.

[@rabomeister]

To be more precise, he just asked "If I do all of the steps exactly, do I have train/valid/test datas -which you should have actually when you want to have a full model- or do I need to do this process on my own?". I hope you can answer it now.

[@pratyakshapokhriyal6405]

Hello ma'am, I am not able to create annotations.txt file using train.csv. I am getting AttributeError: 'NoneType' object has no attribute 'shape'. How to resolve the issue?

[@CodeWithAarohi]

Sorry, Can't help you without check your code. You can mail me at [e-mail]

[@pratyakshapokhriyal6405]

@@CodeWithAarohi Ma'am, I have mailed you my code. Please check.

[@joseandery6699]

@@pratyakshapokhriyal6405 Did you find a solution? I have the same error.

[@TheGeMenge]

Same problem here

[@TheGeMenge]

Fixed, was an index naming problem of my custom dataset. The code block tried to load an image that did not exist. Technically the cv2.imread command should have thrown the error

[@lochan5403]

I'm really grateful to you for the code you've given so far. But could you please give the code to test it too???

[@lintheshwarsakthivel1413]

Hi mam, currently I'm doing a project based on faster RCNN but my dataset is MS COCO what should I do to get working code of Faster RCNN for my dataset????

[@CodeWithAarohi]

You will need to preprocess the dataset to create the necessary input format for Faster R-CNN.

[@lintheshwarsakthivel1413]

Thank you for ur reply mam. I do have an annotation file mam But I cannot understand what should I do with it. Is there any source code for reference or any source for implementing the faster RCNN for my dataset mam. Because I need to implement few shot object detection using Faster RCNN which seems pretty impossible as of now mam. This is very important for me implement mam as my project deadline is nearing@@CodeWithAarohi

[@shaktisinghsonigara9094]

Awesome tutorial.

[@CodeWithAarohi]

thankyou

[@chakrapandasuraj]

Hello Aarohi. A very good explanation was made by you. I have few doubts. Will you please clarify me. 1) I wonder that can we calculate the detected objects percentage in one image? 2) If it is calculated then how to average all the percentages calculated in one image?

[@sachinsrivastava968]

Good explanation 👍

[@CodeWithAarohi]

thankyou

[@nsaidheerajreddy5928]

While creating annotation.txt file, it is showing that 'img' variable is none. Can you help? And also you have a train folder in SAR_faster-rcnn folder. What is that train folder.

[@CodeWithAarohi]

Please provide more details. Not able to understand the error

[@technicalmorujiii2285]

how to download data set....could you please explain run and show it in video, we can follow the same steps, i am getting that error no such file directory found

[@btsofficialnewztv8957]

Where do u execute the 3 commands of python?

[@aditi9568]

height, width = img.shape[:2] AttributeError: 'NoneType' object has no attribute 'shape' What to do ?

[@CodeWithAarohi]

share your code on [e-mail]. I will check

[@CodeWithAarohi]

Create a folder with a name of "train" and put all the training images in this folder ( images of all classes together in this folder, no need to create separate folder for each class.)

[@vinayakkalyanshetti8516]

@@CodeWithAarohi I gave directory path directly to the label folder(which has all the images) yet it is not working!

[@CodeWithAarohi]

@@vinayakkalyanshetti8516 send me your code. And explain what problem you are getting

[@Atulmishra-hs8ch]

@@vinayakkalyanshetti8516 Do something like this: class_name = row['ClassName'] image_path = os.path.join('D:/PyCharm/PyCharmProjects/OIDv4_ToolKit-master/OID/Dataset/train/',class_name+'/') img = cv2.imread(image_path + row['ImageID'])

[@sunainagoyal9121]

While running the code getting the error in training TypeError: Input 'y' of 'Sub' Op has type float32 that does not match type int64 of argument 'x'. Any help??

[@CodeWithAarohi]

Which tensorflow and keras version you are using?

[@sunainagoyal9121]

@@CodeWithAarohi I am using google colab.

[@darshandushing5985]

5:45 where we execute this 3 commands ? Please reply mam

[@CodeWithAarohi]

OPen command prompt. Enter in the folder where your code is. Here you will run the commands.

[@aayushregmi5113]

After the Dataset is prepared, How can we run the data in Faster- RCNN algorithm, Is their any code for it that you can share it would be of great help

[@beingSupraa]

Same problem bro

[@mehnaztabassum1878]

Mam, could you pls share the testing code for Faster RCNN?

[@petermungai3557]

you do the same as you have done to produce the traing annotation file. You change the training data to test data. Hop this is clear

[@chyrilnarvasa2513]

@@petermungai3557 how can i save the training done and proceed to testing dta?

[@moumabenaf940]

​@@petermungai3557how can we use the training annotation file for training faster r cnn please

[@talmus30oz59]

If I were to make my own labeled data set would I need to label all instances of an object in that image. What I mean is... If I have an image with 3 pictures of dogs and I only label one of them as a dog and leave the rest blank will that hurt my model?

[@CodeWithAarohi]

Performance will be poor because sometime you are labeling that object and sometime you are leaving the object

[@laughlive6840]

I don't have an annotation files in csv format, then what should I do?

[@CodeWithAarohi]

You can convert your current dataset into this format.

[@mehnaztabassum1878]

But how could I get the mAP/AP value for the detection?

[@rb1447]

Have you calculated the mAP for this

[@ArunKumar-sg6jf]

img no shape atribute what to do now

[@CodeWithAarohi]

Create a folder with a name of "train" and put all the training images in this folder ( images of all classes together in this folder, no need to create separate folder for each class.)

[@EbookMemories]

@@CodeWithAarohi Sorry, this means that we make all training images in just 1 folder?

[@madhavibisen1545]

@@EbookMemories Have you solved your error bcoz I am also getting the same.

[@GhostRider-du5rh]

@@madhavibisen1545 were u able to solve the error?

[@GhostRider-du5rh]

@@CodeWithAarohi hii, put all the training images in one folder but the annotations.txt file is blank.... train.csv is being created successfully. when i break the code and run, it shows "None" in the entire column for annotations.txt file.

[@safaawan4650]

how to Convert our dataset in the format which fasterrcnn accepts?

[@sanketkadam456]

this video was very helpful ,but can you make a video on custom dataset for face detection

[@CodeWithAarohi]

Sure, Will do soon

[@ammaraikram7291]

no image in my apple orange and switch folders

[@shivamray6689]

mam if i want to make my own dataset in pascal_voc format how can i do it

[@CodeWithAarohi]

Annotate the images in that format. You can use labelImg tool for annotation.

[@Suba-je3wj]

mam,iam doing the snake classification project.Could you please explain how to rain those data sets mam,its very urgent for my project review mam

[@darshandhobi3133]

mam which is very first step after preparing our data format bcz i have data format with .txt and .csv for faster r-cnn.i am confused about ki i have "start training" portion AFTER completion of the vgg,rpn,roi or BEFORE?

[@CodeWithAarohi]

1st step is prepare your data in the format which faster RCNN accepts as described in this video. 2nd step Prepare code for vgg,rpn,roi. And finally train your network

[@SiddharthaBhaskaran]

how to do for a normal dataset just a folder from kaggle

[@mdyounusahamed6668]

for creating the annotation.txt where did you get this? "train_file_path = 'D:/5-deep_learning_with_tensorflow/module 7 cnn/7-faster_r-cnn/SAR_faster-rcnn/train'"

[@_maha_]

Hello, how would you evaluate such model ?

[@sunainagoyal9121]

Is there any video explaining the remaining code of faster RCNN implementation??

[@CodeWithAarohi]

Yes.. check faster rcnn on custom dataset . That will explain the whole process.

[@sunainagoyal9121]

That again explain the dataset part only

[@javlontursunov6527]

Do we need GPU for this work since u r using jupyter notebook on your local machine ?

[@CodeWithAarohi]

You can use GPU or CPU

[@ashzworld8751]

can you please share the testing video of this

[@zubairsk1624]

I'm wondering if you can use ssd with tensorflow v2 please

[@CodeWithAarohi]

I never used it but I think with some changes in the current code we can run with tensorflow2.

[@aryan_01793]

Ma'am, can you please suggest some best papers on RCNN, Fast-RCNN, Faster-RCNN, and YOLO?

[@CodeWithAarohi]

Sorry I don't have references right now.

[@EbookMemories]

Mam, How to create label IsOccluded , IsTruncated, IsGroupOf,IsDepiction,IsInside

[@EbookMemories]

Mam, i have 181 class.. is there problem if I use this code to detect the object?

[@CodeWithAarohi]

HI, Yes you can use this code but you need to define all the 181 classes

[@EbookMemories]

@@CodeWithAarohi Mam.. i wanna asking you... i have my own dataset.. and there is no in open image dataset v6. How i get labelling data that contain Xmin, Xmax, Ymin, Ymax and label IsOccluded , IsTruncated, IsGroupOf,IsDepiction,IsInside? Please reply me.. i wanna working on my final project

[@CodeWithAarohi]

@@EbookMemories 1- Label your images by using annotation tool. Annotation tools are available free on internet.2- Convert those annotated image data in the format which faster rcnn accepts.

[@CodeWithAarohi]

Leave is occluded, isgroupof etc

[@MMSakho]

how about 2 or more class in 1 image mom?

[@binitasajit4394]

Hi Arohi, thanks for a great explanation. But the actual implementation is not done. In your training_fasterrcnn.ipynb, you have written so many function /class like RoiPoolingConv , rpn_layer,classifier_layer, iou functions,loss functions, etc But you have not mentioned how to use these functions. In this program you have concluded at displaying the GT box and the positive anchors. Can ypu please let me know the next steps

[@harinir4608]

same question. did you find any other sources for the implementation of the functions?

[@MrUdaykiranPatnaik]

Hello, Could you please add the last part video for SSD. It is missing.

[@CodeWithAarohi]

Will do that soon

[@anjalinaidukarrotu1431]

i am not getting output with bounding box

[@devalokamsubramanyam8043]

Madam, Very nice videos. Easy to understand by the way you have explained. I need your help. I am unable to get training_faserrcnn Jupiter note book in Github. Pl share the link. Thanks

[@CodeWithAarohi]

github.com/AarohiSingla/Faster-R-CNN-on-Custom-Dataset/blob/main/training_fasterrcnn.ipynb

[@mdwasi7002]

how do we get features from faster rcnn??

[@CodeWithAarohi]

Faster RCNN will detect the objects but features are extracted using any 1 on these Resnet, VGG16 or Feature Pyramid Network (FPN)

[@YENNGUYEN-td6zv]

please tell me how can create custom dataset like you? please!

[@CodeWithAarohi]

You need to have images first and then annotate them using any tools which gives you dataset in the format of csv file

[@YENNGUYEN-td6zv]

@@CodeWithAarohi can you explain how to use this model by taking other pictures and predict the labels on the images? because I don't find anything to use this faster rccn trained model to make our own predictions. Kindly help me in this regard. Pleaseee!

[@soravsingla8782]

Amazing

[@CodeWithAarohi]

Thanks

[@CreationAI-se4vd]

Hi , What is base path in the code?

[@AHMADKELIX]

permission to learn mom. thank you

[@seetharamnageshappe5013]

Thank you for video.But when I am training a model of MAskRcnn or FasterRCnn) using keras iam getting a lot of errors like keras.engine as no layer even with tensorflow 1.14.1 and keras 2.3.0

[@CodeWithAarohi]

which version of h5py you are using?

[@seetharamnageshappe5013]

@@CodeWithAarohi madam will you give the list of technologies along with versions to rum the mask rcnn

[@CodeWithAarohi]

@@seetharamnageshappe5013 only tensorflow, keras and h5py version is important

[@seetharamnageshappe5013]

@@CodeWithAarohi Requirement already satisfied: h5py in /usr/local/lib/python3.7/dist-packages (3.1.0) Requirement already satisfied: cached-property; python_version < "3.8" in /usr/local/lib/python3.7/dist-packages (from h5py) (1.5.2) Requirement already satisfied: numpy>=1.14.5; python_version == "3.7" in /usr/local/lib/python3.7/dist-packages (from h5py) (1.19.5)

[@seetharamappe7899]

@@seetharamnageshappe5013 I am using custom data and want to label for fastercnn. which tool should i need to use?

[@anjalinaidukarrotu1431]

link for continuation video pls

[@ArunKumar-sg6jf]

how to use video for faster rcnn

[@CodeWithAarohi]

Convert your videos into frames and then apply faster rcnn

[@leenavig3441]

where to run the first three commands???

[@CodeWithAarohi]

In command prompt

[@taniasultana5865]

Thanks for your good works. For the custom dataset of 2 classes (with car and without car), how can I make the train/test/validation annotation_bbox.csv file? Could you please guidelines me?

[@CodeWithAarohi]

Use OID to collect data for car class. and for without car class, you need to download images from internet and then divide them into train and test and after that use any annotation tool to annotate your images and after that convert your data into the format which fasterRCNN accepts

[@taniasultana5865]

Thank you so much. Could you please upload the code for the test part?

[@debupatra2333]

Ma'am, can you please upload the requirements.txt file also, I want info about the versions of the libraries. Great video, thank you

[@CodeWithAarohi]

tensorflow==1.15.0, Keras==2.2.5, numpy==1.19.5,

[@cricketdiehardfan9306]

Hello Aarohi, I have a trained dataset which is used for the yolo model and i need to train the same dataset for faster rcnn. can you suggest me the steps to train faster rcnn using this dataset.

[@CodeWithAarohi]

Convert your dataset in the format which fasterrcnn accepts and then follow the steps which I have mentioned in this video

[@alvinl1234]

thanks!

[@CodeWithAarohi]

Welcome!

[@CodeWithAarohi]

Welcome!

[@ArunKumar-sg6jf]

make full tutorail for computer vsion

[@CodeWithAarohi]

Sure, Will do

[@hassanmehmood8934]

Hello Mam its a great work you have done. @Code With Aarohi Mam can you explain how to use this model by taking other pictures and predict the labels on the images? because I don,t find anything to use this faster rccn trained model to make our own predictions. Kindly help me in this regard. Thanks

[@CodeWithAarohi]

I will try to do it soon but right now busy with my Project work

[@hassanmehmood8934]

@@CodeWithAarohi Thanks for considering this. ok sure when ever you have time then make a tutorial on that. it will be very appreciative.

[@ArunKumar-sg6jf]

MAKE VIDEO FOR YOLO V4 madam

[@CodeWithAarohi]

Sure. Will do it soon

[@ArunKumar-sg6jf]

👌👌👍👍

[@Atulmishra-hs8ch]

Mailed you regarding an improvement. Kindly check.

[@CodeWithAarohi]

Ok

[@developmentsproject9188]

ver vague, following your step instead of straight forward.

[@israrahmed7951]

Hello mam, Could you Please check out your mailbox I have sent you mail

[@CodeWithAarohi]

Checked and replied

[@israrahmed7951]

@@CodeWithAarohi could you please tell me the version of mxnet

[@mythray9997]

cute

[@akshayv6982]

For those who are running on infinite loop at X, Y, image_data, debug_img, debug_num_pos = next(data_gen_train) add this below code before calc_rpn : def union(au, bu, area_intersection): area_a = (au[2] - au[0]) * (au[3] - au[1]) area_b = (bu[2] - bu[0]) * (bu[3] - bu[1]) area_union = area_a + area_b - area_intersection return area_union def intersection(ai, bi): x = max(ai[0], bi[0]) y = max(ai[1], bi[1]) w = min(ai[2], bi[2]) - x h = min(ai[3], bi[3]) - y if w < 0 or h < 0: return 0 return w*h def iou(a, b): # a and b should be (x1,y1,x2,y2) if a[0] >= a[2] or a[1] >= a[3] or b[0] >= b[2] or b[1] >= b[3]: return 0.0 area_i = intersection(a, b) area_u = union(a, b, area_i) return float(area_i) / float(area_u + 1e-6)

[@petermungai5508]

Hi mam I send something in your email can you check it plz?