Image classification with Python and Scikit learn | Computer vision tutorial

[@ComputerVisionEngineer]

Did you enjoy this video? Try my premium courses! 😃🙌😊 ● Hands-On Computer Vision in the Cloud: Building an AWS-based Real Time Number Plate Recognition System bit.ly/3RXrE1Y ● End-To-End Computer Vision: Build and Deploy a Video Summarization API bit.ly/3tyQX0M ● Computer Vision on Edge: Real Time Number Plate Recognition on an Edge Device bit.ly/4dYodA7 ● Machine Learning Entrepreneur: How to start your entrepreneurial journey as a freelancer and content creator bit.ly/4bFLeaC Learn to create AI-based prototypes in the Computer Vision School! www.computervision.school 😃🚀🎓

[@valnuke]

you're the best Felipe! short videos straight to the point, great explanations

[@ComputerVisionEngineer]

Thank you for your support! 😃💪

[@John-xi2im]

ultimate teaching style, this was the best computer vision introduction tutorial I have ever seen on youtube as it was very easy to understand , thanks a lot Philip!!

[@ComputerVisionEngineer]

You are welcome! Thank you for your support! 😃🙌

[@philippe6724]

Philipp did an excellent job at explaining all the details of sklearn. Thank you for sharing your knowledge-)

[@imadsaddik]

The way you explained this topic is amazing, I enjoyed watching the video and I didn't even feel the length of it.

[@ComputerVisionEngineer]

Hey Imad, thank you! I am so glad you enjoyed it! 😄💪

[@nazarmerza1]

This an amazing tutorial. Great job Felipe!

[@kenatsharaby]

Great stuf!

[@ComputerVisionEngineer]

😃 Thank you so much for your support!! 🙌🙌

[@kenatsharaby]

@@ComputerVisionEngineer Your' welcome. Question: how can I contact you privately to hire you for an ai project I am running?

[@ComputerVisionEngineer]

You are welcome to tell me about your project through discord. The link is in the about section.

[@CarlosAlejandro-g8h]

Thank you, very clear and easy to follow tutorial

[@the.patricklong]

AWESOME! Made my first working ML model. Thank you so much!!!!

[@ComputerVisionEngineer]

😃 That is so cool!! You are welcome! 🙌

[@matthiasandreas6549]

Thank you so much for this Video. Makes fun and have learning effect.

[@ComputerVisionEngineer]

😃 Glad you enjoyed it! 🙌

[@Akios281]

Thank you very much, Felipe, for these very clear explanations. Until now, I was making predictions for numerical values. Before watching your videos, I thought image classification was more complicated (even though this is a relatively straightforward use case). I have just one perhaps silly question: How did you create the image captures for the dataset? Was it done manually, or did you use specific software for this somewhat tedious task? Thanks again!

[@ComputerVisionEngineer]

Hi, I took the parking lot video I show in the tutorial, I created a binary mask with the location of all the parking slots, then got all the connected components using opencv, then cropped the bboxes from the video frames. 🙌

[@deco-f7y]

Your tutorial is very cool, thank you very much for helping me with your tutorial

[@ComputerVisionEngineer]

😊 You are welcome! 🙌

[@abdvi4423]

I love the explanation and you helped me so much!

[@ComputerVisionEngineer]

You are welcome! Glad you found it helpful! 😃🙌

[@hamzawi2752]

Thank you so much for your time, you did a great job.

[@ComputerVisionEngineer]

Thank you for your support, Hamzawi! 🙂🙌

[@AbhishekRaj-ig3gk]

Very helpful ❤

[@ComputerVisionEngineer]

😊 Thank you, Abishek! I am glad you find it useful! 🙌

[@alejandromontiel3616]

Thanks for your content. I an working on a CNN to train and predict rock masses Geological strength index (it goes from 0 to 90 each 5, and it is an index that tells us how fractured a rock mass is) , so I took some pictures of rocks masses and I labeled them with their GSI index. So, Any recommendations for me? I mean I want my CNN to indentify patterns, basically join sets, so and predict rocks masses GSI. Is it a doable task for any machine learning model? Thanks from New Mexico, USA.

[@ComputerVisionEngineer]

Hi, it's hard to give you a detailed recommendation without knowing more about your project, but make sure your data is accurate and well-labeled, reflecting the categories you're working with. Ensure you have a diverse set of images and aim for a few thousand labeled examples. Once your data is solid, a simple CNN should work well, and you can fine-tune it from there. Good luck! 🙌

[@guilhermegodoi2336]

Amazing job Felipe, you are helping me so much on my studies! I'm from Brazil and I will definitely use your videos to help on my completion of course work

[@ComputerVisionEngineer]

Hey Guilherme! It makes me so happy the videos are helping you! 😃😃 Thank you for watching! 🤓💪

[@shivammehta9858]

A great session by you sir , sir pls also share the code to input a photo in it and put it in the trained model to show the results of analysation

[@ComputerVisionEngineer]

Do you mean the code to make predictions on individual images? I will try to add that code soon.

[@HaxeHere]

Thanks a lot for your content Felipe, you're a great teacher! I'm new to machine learning so there's one thing that confuses me: how can I now utilize the the model that I've created? I understand that Pickle transforms the data into byte form and that it can be read somehow, but could you give me an example of what to now do with this data?

[@ComputerVisionEngineer]

Sure! Take a look at this function, here is an example of how I use the same image classifier I trained in this video to make predictions github.com/computervisioneng/parking-space-counter/blob/master/util.py#L14 Please mind I am loading the model in line 11. 💪💪

[@HaxeHere]

Thank you!

[@TatianaMaskaron]

hi, qq how did you split the images to empty and not empty? I saw the other video where you cropped images, but how did you separate 6K+ images into empty and not empty, its not clear.

[@ComputerVisionEngineer]

The opencv function I used to get all connected components provides an id for every component, so I looked at the ids of empty slots and non empty slots and save them into different directories.

[@TatianaMaskaron]

@@ComputerVisionEngineer i see, thank you. Yes I also looked at the other videos you have and realized how its done. Thanks for responding. Btw I pinged you on linedin for a project. I did not see reply, assuming you are not logging in to that often. Would you be interested in a paid consultation?

[@ComputerVisionEngineer]

@@TatianaMaskaron thank you so much for reaching out regarding a paid consultation, I do work as a freelance contractor, but I am a little too busy right now with other projects. Yeah I don't use linkedin that much, but you can reach out to me through mail, discord, Patreon, or the comments in my videos, I try to answer as soon as possible. 🙌

[@TatianaMaskaron]

@@ComputerVisionEngineer Filippe, if you can pass me your email that would be great. I don't know how to reach out on here without using comments.

[@abdulazizkadirbaev604]

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (6090,) + inhomogeneous part. what to do?

[@datascientistonline12]

Hi, great tutorial! I seem to be stuck on a problem when I execute the "grid_search.fit(x_train, y_train)" part of the code. This is the issue "FitFailedWarning: Estimator fit failed. The score on this train-test partition for these parameters will be set to nan. Details: TypeError: only size-1 arrays can be converted to Python scalars". Do you know how to fix this? I have ensured to use the correct versions (scikit learn, scikit image, and numpy) as shown in video, and followed the video step by step. This is the code for importing images i have (i had to add dtype=object when converting lists to numpy array as it was giving me visibleDeprecationWarning): for category_idx, category in enumerate(categories): for file in os.listdir(os.path.join(input_dir, category)): img_path = os.path.join(input_dir, category, file) img = imread(img_path) print(len(data)) img = resize(img, (200, 200)) data.append(img.flatten()) labels.append(category_idx) data = np.asarray(data, dtype=object) labels = np.asarray(labels, dtype=object). Finally, do you have an instagram or other site to discuss this issue, if possible?

[@ComputerVisionEngineer]

Hey, what is your Python version and your OS? Many things could be going on. I would start by making sure the data and labels are being loaded correctly. You could print 'data' and 'labels' and see if they make sense.

[@FahadKiani1]

Felipe the Legend 👏🏽

[@ComputerVisionEngineer]

lol, thank you for your support! 🙌

[@alephzahir2780]

Hello , can u help me to know how go use my own data but i want to put an image and the code classify the image belong to any class , thank you

[@cattnation6257]

i need your advice i have task where i need to find avarage pixal value of x-ray image object in x-ray image i have many object like pen paper palstic and so on and each intesity is different some are overlap some have no clear edegs and mixed edges can i detect or draw a boundry box around each object is it possible without ml model i am newbew in this field specially in image processing can it can be done by image process alone can i segement object if intesity of each object is different and edges are not so clear please help me i am single develpoer in my team

[@andres_camarillo]

Great video!

[@ComputerVisionEngineer]

Thank you! 😃🙌

[@cenizamaricar7911]

Hello Sir, can we use this scikit in egg sexing for our incubator?

[@yanhu2011]

Great video. Is there a follow up on this to show how you can use the classifier and display the annotated image with green box and red box representing occupied and unoccupied spots in a parking lot image?

[@ComputerVisionEngineer]

Thank you! Sure, visualization is covered in the parking spot detector and counter video! kzbin.info/www/bejne/fF6baWeAZ9Khhbs

[@abdullahazizhussain8117]

amazing tutorial, thanks! I had a question about the model.p file, is there some way I can open it and view its contents, I want to know how it classified, or maybe to see if the task was done as I wanted it to be done.

[@ComputerVisionEngineer]

For models trained with scikit learn, in some cases is possible to look at the model weights or some of its internal parameters. But in general I would say model.p is a binary file and it only makes sense in scikit learn, there is not an easy way to 'open it and view its content'. 🙌

[@itsthemeg]

Could you make another video when you test out this classifier?. Like host this online and let the user upload images of parking spaces and the program outputs 'empty' or 'not empty'. Thanks

[@ComputerVisionEngineer]

I will try to. 🙌

[@YigalBZ]

Thank you for a great clear video. I am going to run it through. Yet something is missing for me: I would like to train a set of specific toy car recognition. For that I need to take pictures of the toy car , and pictures of the floor/table (where the car can be placed) for the "empty' part. Can you recommend: 1) how many pictures of each? 2) What resolution? does it matter? 3) Any other guide line you would recommend?

[@ComputerVisionEngineer]

About the number of pictures, you should be able to achieve a high level of accuracy with a 'small' dataset; try with 200 images in each class. The resolution may not be that critical, but try to make it as high as possible. 🙌

[@mayankmalhotra3523]

Hi Felipe, Actually I wanted to ask that is this model build using pretrained model or not? and If it is build on pre trained model can you suggest me a video of yours where I can learn building a model with using a pretrained model

[@ComputerVisionEngineer]

Hi, I build this model using scikit learn, I am not using a pre trained model. 🙌

[@mayankmalhotra3523]

also i wanted to know about your video named as sign language detection with scikit learn. The thing is in that video the sign detection one of yours are you using any pre trained model?

[@ComputerVisionEngineer]

@@mayankmalhotra3523 yes in that video the landmark detection is with a pre trained model.

[@mayankmalhotra3523]

@@ComputerVisionEngineer hey felipe i loved the tutorial but if i wanna see how the model recognises the parking spaces in a video and its real time working ? actually i wanted to see the output of it

[@rakeshkumarrout2629]

This is really anazing.can you make a video on how to apply GAN and use it or make it from scartch?

[@ComputerVisionEngineer]

Hey Rakesh, making a video about GANs is a very good idea! I will definitely keep it in mind for future videos! 😄

[@informationnet]

It helped a lot for me.

[@ComputerVisionEngineer]

Awesome! I am happy you found it helpful! 😄🙌

[@mirtanvirahmed3868]

amazing tutorial indeed! ❤️ 👌 can you suggest me some resources where i can find some free data for training models?

[@muhammadahmadessa9584]

Great video, could I use this same code to classify more than 2 categories? or does it only classify in binary, like e.g empty not empty.

[@ComputerVisionEngineer]

Hey, thank you! Yes, you can use it to classify more than 2 categories! 😃🙌

[@MrThefakiri]

You are use category_index for labels ?? I think the labels is category

[@saketolfa]

Thumbs up for this video. I just noticed a small mistake . I think you accidentally switched the parameters for the accuracy_score function. It should be accuracy_score(y_test, y_prediction) instead of accuracy_score(y_prediction, y_test).

[@ComputerVisionEngineer]

Hey, thank you for your contribution! Although the order of input parameters is not crucial at computing the accuracy, the function outputs the same result 💪🙌

[@江東諺-o1q]

Could you release the data of empty and not empty for us? Thank you very much for your amazing tutorials on OpenCV and Machine Learning. It is beneficial for me to construct my own project!

[@ComputerVisionEngineer]

Sure! I am happy to help you in your machine learning journey! 😃🙌 Here you go: drive.google.com/file/d/11DyZ165lZGzULEZSQNofyy9A8xaYgFJ2/view?usp=share_link

[@vivekpatel009]

Amazing tutorial !!, Just a question, there are many classifiers in sklearn. Why did you choose SVC specifically ?

[@ComputerVisionEngineer]

No reason. Given the data I used in this project, I think pretty much any other classifier would have had a similar performance. 💪💪

[@nwbstplyr]

cool video! can i possibly import this model to like a huskylens? thanks!

[@ComputerVisionEngineer]

Hi, if you can execute Python and Scikit learn, then most likely yes! 🙌

[@nwbstplyr]

@@ComputerVisionEngineer sounds great, trying to come up with a plan to computer vision something either its a good object or a bad one depending on some criteria

[@hassanmohamud6415]

hi um i dont know why i dont know how to get data i did get from desc but i still didnt know if you could tell me that would really help

[@itsthemeg]

Can this prediction model be used for any set of data?

[@ComputerVisionEngineer]

Sure, it is an image classifier. There are other approaches that may yield a higher accuracy, though.

[@dipamsarkar2403]

Sir, I have facing some error while training the classifier. It says : ValueError: Found input variables with inconsistent numbers of samples: [4872, 1218] It would be of great help if you solve this problem🙏

[@saifeldenkhaled9716]

it'd be better if you put the snippet of code so we can check it out with you

[@MrAiblack]

I have a task to classify cat and dog so i tried the same way you do it but the accuracy its very low i used gray scale to minimize the complicity but didont work , i want to used like this approach if you has any advice

[@ComputerVisionEngineer]

Try changing the size you are resizing the images to, line 23 from file main.py. What is the size of your images?

[@MrAiblack]

​@@ComputerVisionEngineer I putted the size to (50,50) To make all image to one size but the image become a worst

[@ComputerVisionEngineer]

@@MrAiblack by increasing the size in line 23 performance should go up. Nevertheless, this may not be the best approach for this type of task. I would recommend you to use feature extraction before applying a scikit learn classifier. I may do a video about it soon. 💪

[@MrAiblack]

@@ComputerVisionEngineer Ok ,Thank you vary much,I will be waiting

[@amanPatel-mw7ly]

not able to download dataset

[@Matlab-ku3ns]

where do i get model.p file?? it is not available in code nor data link given in description

[@ComputerVisionEngineer]

Hey, in order to create the model.p file you need to execute 'main.py' from the project's repository. 🙌

[@group4661]

Hello, can you help me to come up with python code that classify duck eggs such as balot, penoy, abnoy according to it's classification?

[@ComputerVisionEngineer]

Hey, that sounds like a cool project! As long as the different classes are visually distinct, you can classify them with an image classifier! I suggest you start with the data collection; collect data from all the different categories you wish to classify. Once you have the data, the Python code would be very similar as the one I used in this video! 😃💪🦆

[@silentt2143]

If I have more that 2 categories of images, can I name “x_train” and “x_test” whatever I want? For example “car_train”, “car_test”, “bus_train”, “bus_test”, and “limo_train”, “limo_test”?

[@ComputerVisionEngineer]

Hey, you can certainly name your variables whatever you want. But, remember the category is already encoded in the labels variable. For example if you are building a vehicle classifier and the categories are car, bus and limo. Then each data point will be labeled as 0 (car), 1(bus) or 2 (limo).

[@silentt2143]

@@ComputerVisionEngineer ok cool. Thank you!

[@SleepySoundscapes777]

hey , great Tutorial thank you , but how do i use it now to classify new images ?

[@ComputerVisionEngineer]

Hey, oh I see the inference of individual images is not covered in this tutorial. Take a look at this video kzbin.info/www/bejne/g3umhJ2OmtWUd68, in this video I also train a classifier using scikit learn and I show you how to inference new data. 🙌

[@SleepySoundscapes777]

@@ComputerVisionEngineer thank you so much

[@BeeLiBahLa]

Im confused, which part of the code tells it that the picture has a car or not?

[@arbababdurrab5702]

do you find out??? if yes then plzz also tell me

[@all_rounder_786_]

Hello Sir, i used the same code, but its not giving me the output result, like don't print the accuracy... kindly help! thats what it is showing in the notification area. Shared indexes are downloaded for Python packages in 1 min, 12 sec (20.48 MB)

[@ComputerVisionEngineer]

Hey, I just tested the code and everything seems to be working well on my end! That msg seems to be related to Pycharm, maybe you could try to execute from another IDE or from the command line?

[@yagmur3718]

I'm experiencing the same, any suggestions for a solution?

[@peterMwaura-rh3nm]

Thanks. Is there a way I can convert the model to a tflite

[@ComputerVisionEngineer]

Hey, the image classifier we used in this video is from scikit learn library, it is not possible to convert it to tflite. 🙌

[@peterMwaura-rh3nm]

@@ComputerVisionEngineer okay thanks a lot. I will be on the lookout for Keras and tensor flow. Thanks

[@tharndu3387]

Why don't you annotate the images using bounding box or any methoda ? Is it not compulsory ? 🥲

[@ComputerVisionEngineer]

Hey, given this is an image classifier the type of annotation we need is only the image category. 💪💪

[@indointanchannel]

Thank you, sir. Could you share the dataset used?

[@ComputerVisionEngineer]

Sure! I have added a link to the data in the description of this video! 😃

[@indointanchannel]

@@ComputerVisionEngineer Thanks, brother

[@F_F_CH]

Ciao Fillippe, I have question about this profession in 2024 and want to know your opinion. Do I need to study any course in online platforms to become a good on it and of course finde a job(take certificate for CV). And is it easy to find a job as Comp.Vision engineer nowadays(2024 and maybe future). What can you advise me to become an engineer as you (online courses, books ect...) P.S. I really appreciate your contents, thank you for your time and effort. Hope to become a professional com.vision engineer and send you a gift from my 1st salary(but now i dont know, how to become it)

[@ComputerVisionEngineer]

Hi, take a look at the computer vision roadmap video, there are links to youtube videos and resources you can use to learn computer vision engineering. 🙌

[@horyantan5899]

how if I need to use CNN network? Example AlexNet?

[@ComputerVisionEngineer]

Hey, I may make another video in the future on creating an image classifier based on deep learning, like AlexNet. 😃💪

[@niklaskeller8465]

How do you build up your trainig data? You won‘t have decided for 6000 pictures whether they are empty or not? Did you use kind of a „step by step“ classification? So that with lets say with 50 pictures you built up a classifier, that builds up further (and of course checked it then whether it worked properly? Thanks for your videos!

[@ComputerVisionEngineer]

Hey Niklas! That is a good question. The data I used in this video comes from a previous video where I built a parking space detector and counter. In that other video I got the location of all parking slots by getting all the connected components of a binary mask. The function I used for getting the connected components got me not only the bbox but also an id for that specific location. Then, as most of the spaces are either empty or not-empty during the entire video, I just selected a few empty spaces (and a few not-empty spaces) and cropped them out of the video in all frames. 💪💪

[@tharndu3387]

When i was trying to convert data and labels into numpy array, i got this error😢 ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (84,) + inhomogeneous part. What is this ? Thanks for the amazing tutorial ❤

[@ComputerVisionEngineer]

Hey, what is your OS and your Python version? Have you installed the requirements listed in the requirements.txt file?

[@tharndu3387]

@@ComputerVisionEngineer No sir, first i used a data set that i have created, when i load your data set, now it is working 😍 But I want to know what is that error ? How it occurred 🥲

[@yacinerouizi844]

thank you felipe

[@ComputerVisionEngineer]

You are welcome! 😃

[@EndahSugiatri]

how to deploy your pickle model on android app?

[@ComputerVisionEngineer]

Hey Endah, there are different ways in which you can deploy the model we created in Scikit learn in an Android app. If you are looking to run the model in the mobile device you could consider the Python library sklearn2onnx in order to convert the Scikit learn model into an onnx file. 😃💪

[@EndahSugiatri]

@@ComputerVisionEngineer do you have any reference to change the model? I had trouble finding the reference, I used this classification method for my thesis and my deadline is only 2 days to be able to run this program. I hope you can help me, thank you

[@ComputerVisionEngineer]

Try to Google "deploy scikit learn model android app" there are a few articles on how to do it, you may find the solution in one of those articles. 😃

[@zilefacebenezernwetlawung3200]

thanks very much can the score be 100%

[@ahmedsmair4416]

why music ?

[@erfanalaei8784]

i love you

[@mayankrohilla4105]

Why am I getting an accuracy score of 1.0 xD

[@salvadorliebana6079]

Que crack

[@bisyrim1030]

I have 100.0% of samples were correctly classified 😁

[@ComputerVisionEngineer]

Awesome! 😃🙌