A fun, hands-on deep learning project for beginners, students, and hobbyists

Today’s blog post is a “bonus tutorial” in our most recent series on building a complete, end-to-end deep learning application:

In this series we have been fulfilling a childhood dream of mine: to build a Pokedex.

A Pokedex is a fictional device from the world of Pokemon (I was/still am a huge Pokemon nerd) and allows the end user to:

  1. Point it at a Pokemon (animal-like creatures), presumably using some sort of camera
  2. And automatically identify the Pokemon, providing detailed information on the creature

You can therefore think of a Pokedex as a smart phone application that (1) accesses your camera and (2) recognizes animals/creatures in real-time.

In order to identify Pokemon we trained a Convolutional Neural Network using Keras — this model is capable of correctly identifying Pokemon in images and video streams.

The model was then deployed to a mobile app using Keras, CoreML, and iOS to create an actual “Pokedex app”.

But why stop there?

Long-time readers of the PyImageSearch blog know that I love the Raspberry Pi…

…and I couldn’t help myself from building an actual Pokedex device using a:

  • Raspberry Pi
  • Camera module
  • 7″ touch screen

This series has certainly been a fun, nostalgic project — thank you for taking this journey with me.

To learn more about this fun deep learning project, and run deep learning models in real-time on the Raspberry Pi, just keep reading!

Looking for the source code to this post?
Jump right to the downloads section.

A fun, hands-on deep learning project for beginners, students, and hobbyists

In today’s blog post you are going to learn how to build a complete end-to-end deep learning project on the Raspberry Pi.

This project is excellent for beginners, students, and hobbyists interested in applying deep learning to their own applications.

We’ll start off today by reviewing the hardware I used to build this project.

All of this hardware is optional, including the Raspberry Pi itself (you could run this code on your laptop/desktop provided you have a built-in webcam or USB camera). Feel free to use all or none of the hardware detailed today.

From there we’ll review the directory structure for our project.

We’ll then implement our real-time deep learning application on the Raspberry Pi using Python and Keras.

The model we’ll be using here today has already been trained. To learn how we trained our Keras model, please refer to this blog post.

Finally, we’ll deploy our real-time deep learning application to the Raspberry Pi. I have provided a number of video and GIF animations showcasing our Raspberry Pi + deep learning application.

Raspberry Pi hardware and components

To build our Pokedex deep learning application I am using the following components:

Again, all of these components, including the Raspberry Pi, are totally optional. You could easily run this program with your laptop and webcam.

I simply decided to use the Raspberry Pi as I like the hardware and enjoy working with it.

To get started, you’ll want to lay out both your Raspberry Pi and touch screen:

Figure 1: My Raspberry Pi + PiCamera (top) module and a 7″ touchscreen (bottom) laid out before assembly.

On the top I have my Raspberry Pi and on the bottom is the touchscreen.

Notice how I already have the camera module attached to the Raspberry Pi.

The Raspberry Pi screws on top of the touchscreen adapter board:

Figure 2: I’ve attached a Raspberry Pi 3 to a Pi-compatible 7″ touchscreen for my deep learning project.

I was incredibly impressed with the Pi Foundation touchscreen. Assembling it was a breeze and it worked perfectly without any special driver updates or installs.

From there I used a strip of electrical tape to attach the Raspberry Pi camera module to the back of the touch screen:

Figure 3: A deep learning Raspberry Pi project can be made with a Raspberry Pi 3, PiCamera module, and 7″ touchscreen. Shown are the Pi and PiCamera attached to the back of the touchscreen.

Notice that the camera is facing towards us in this orientation (Figure 3). This is the opposite orientation of a webcam — we’re not using our camera to take selfies. Instead, our camera will be used to capture images of Pokemon that are in front of us.

I then attached the power supply and booted up the Pi. Here you can see me launching the pokedex.py  script (which we’ll create later in this guide):

Figure 4: My deep learning computer vision project with the Raspberry Pi is ready to go. Here you can see I’m launching the application from the terminal.

Again, notice that the camera is facing away from us while the screen is facing towards us in this orientation (Figure 4). Don’t make the mistake of attaching the camera to face you (as I did…twice).

Once the script is up and running you will see the Pokedex user interface:

Figure 5: This week’s Raspberry Pi deep learning project is a Pokemon Pokedex — as you can see, the application is loaded and we’re ready to catch them all!

To learn how to configure your system, and build the actual Pokedex itself, refer to the rest of the tutorial.

Configuring your Raspberry Pi for deep learning

Figure 6: We’re going to use Keras with the TensorFlow backend to implement the Raspberry Pi deep learning project (a Pokedex that can identify five common Pokemon species at the touch of a finger).

In this post we will be running a Keras deep learning model on the Raspberry Pi using the TensorFlow backend.

Unfortunately, installing TensorFlow on the Raspberry Pi isn’t as easy as pip install tensorflow .

The gist is that you’ll need to:

  1. Utilize Python 2.7 instead of Python 3 (which is a bit of a bummer but acceptable since this is one-off project)
  2. Download a special TensorFlow wheel ( .whl ) file that has been prepared for the Raspberry Pi (I’ve included this file in the “Downloads” section of this blog post)
  3. Install the required TensorFlow + Keras system dependencies (such as HDF5)
  4. Install the required Python dependencies
  5. Install TensorFlow 1.1.0 via the special .whl  file (credit to GitHub user samjabrahams)
  6. And finally install Keras version 2.1.5 via pip (for compatibility with TensorFlow version 1.1.0)

The process is fairly straightforward, but is too lengthy to include in this blog post.

If you would like to install Keras and TensorFlow on your Raspberry Pi please refer to the “How do I install TensorFlow and Keras on the Raspberry Pi?” section of this blog post.

As far as your OpenCV install goes, I suggest you follow this tutorial to ensure you install a NEON + VPF3 optimized OpenCV.

You will also need my imutils package, which you can install via pip:

In the remainder of this post I will make the assumption that either (1) you have properly configured your Raspberry Pi for Keras + TensorFlow, or (2) you are not using the Raspberry Pi and your system is properly configured.

Training our deep learning model

Figure 7: Training and validation loss/accuracy plot for a Pokedex deep learning classifier trained with Keras.

The Convolutional Neural Network we are using for image classification in this post has already been trained on a dataset of Pokemon (5 classes over 1,191 images).

We were able to obtain 96.84% classification accuracy on our training set and 97.07% accuracy on the testing set.

I have included the trained model in the “Downloads” section of this blog post.

To learn how to train the model, be sure to visit the previous tutorial.

Deep learning project structure

Before we get started coding up our real-time deep learning application on the Raspberry Pi, let’s first examine the project and directory structure:

In the project directory, there is a folder named assets  and a Python script named  pokedex.py .

The pokedex.py  file is our application/driver script which will:

  1. Load the model
  2. Kick off the GUI and deep learning inference process

Residing in the assets  directory, the lb.pickle  file (an instance of scikit-learn’s LabelBinarizer  class) contains our class labels. This file was generated in our previous post on Keras + Convolutional Neural Networks.

Let’s go ahead and deserialize this file and examine the class labels with Python 3 and pickle , the Python version our model was trained on. We’ll need to manually copy and paste these labels into our pokdex.py  file to avoid any Python 2.7 vs. Python 3 serialization issues:

You can leave your terminal open or refer back to this block so you can copy/paste the class labels into our next script.

Let’s examine the remaining files in the assets  directory.

The pokedex.model  file is our trained Keras image classifier — our pokedex.py  script will load our Keras clssifier from disk and apply it to input images/frames.

The pokemon_db.json  file is a simple JSON file containing the height and weight of each of the Pokemon which our Keras model was trained to recognize:

Our pokedex.py  script will load this database from disk during initialization and then display the appropriate height and weight for each Pokemon when classified.

You can think of the pokemon_db.json  file was the “encyclopedia” portion of the project. You can add more information to the encyclopedia, remove details, or swap out the JSON file for a sqlite implementation — the choice is up to you, have fun with it!

Finally, we have two images, the first of which is pokedex_bg.png :

Figure 8: The Pokedex background image for our deep learning project. Credit goes to Game Trader USA.

This image (found on Game Trader USA‘s site) is the “user interface” to our deep learning application.

The black screen-like region on the left will show the real-time video feed from the Raspberry Pi.

On the right, in the green area, relevant information on the Pokemon including species, height, and width will be displayed upon classification.

We then have pokedex_mask.png :

Figure 9: The mask allows us to place images overlaid on the background GUI image. Masking is a topic covered in Chapter 6 of my very first book, Practical Python and OpenCV + Case Studies.

This mask is a binary image and contains only two pixel values: 0 (black) and 255 (white).

When creating our user interface, we’ll mask our input video stream to ensure it fits nicely into the pokedex_bg.png  (complete with the rounded corners!).

Implementing real-time deep learning on the Raspberry Pi

Now that we have reviewed our project structure, let’s go ahead and get started.

Open up pokedex.py  and insert the following code:

Lines 2-10 handle importing packages for this project.

Notably, we’ll be using Keras’ load_model  to load our pre-trained deep learning model (upon which we’ll make predictions). We’ll be utilizing VideoStream  from my very own imutils package to handle our live video stream.

Each of the requirements are installable via pip except for OpenCV and TensorFlow (a backend for Keras). Refer to “Configuring your Raspberry Pi for deep learning” section above for any package installs.

Moving on, let’s review our CONFIG  dictionary:

Here we’ve defined a configuration dictionary.

Our first key,  "model_path" , contains the path to our input deep learning model. Today we’re using the pokedex.model  (trained two weeks ago) which resides in the assets directory.

Next, "labels"  is our set of class labels which our Keras deep learning model was trained to recognize in our previous post.

I actually pasted these values from the terminal Python interpreter shown above for readability and compatibility (rather than loading the pickle file and inserting into the dictionary programmatically).

How come?

Python 2.7 and Python 3 handle pickle files differently (try to deserialize a Python 3 pickle file in a Python 2.7 environment and you’ll see).

To overcome this Python 2.7 and and Python 3 issue, I simply hardcoded the dictionary in the script. An alternative would be loading a text or JSON file of class labels as well.

Let’s check out the rest of the CONFIG  dictionary:

In this block, we’ve have four more keys + values:

  1. "db_path" : The path to the Pokemon JSON database which contains information we’ll be displaying about the little critters.
  2. "display_for" : The number of frames our Pokemon information will be displayed for upon successful classification. I’ve set this value to 240 frames.
  3. "pdx_bg" : Our Pokedex background GUI image path.
  4. "pdx_mask" : The GUI’s associated mask image path.

The remaining keys + values in our configuration dictionary are a selection of image coordinates and the text color:

To determine the exact (x, y)-coordinates of where the video stream for the user interface should be displayed, I opened up pokedex_bg.png  in Photoshop and examined the coordinates.

I found that (25, 125)  is the appropriate location.

You can use whatever tool you are familiar with, I just use Photoshop regularly and am familiar with the interface.

Similarly, Lines 40-45 define the (x, y)-coordinates for where the Pokemon’s name, height, and weight information will be displayed (again, these coordinates were also determined by examining pokedex_bg.png  in Photoshop).

The value associated with the "pdx_color"  key is the RGB tuple of the color we are going to use when drawing the Pokemon name, height, and weight on the user interface. We define this color as RGB tuple and then reverse it since OpenCV expects colors to be passed in using BGR ordering.

Going forward, be sure to refer to the CONFIG  dictionary as needed.

Next, I’ll initialize some important variables and explain what each represents:

The first variable, frame , is the current frame captured from the video stream.

Following is clicked  — a boolean indicating if the screen was tapped (it is a touch screen after all, but your mouse will also work). In a previous PyImageSearch tutorial, I demonstrated how to capture mouse click events with OpenCV — we’ll be using a similar method here. When a user clicks/taps in our application we’ll classify the input frame and display the results for a set amount of time (240 frames in my case).

The counter  variable holds a count of the number of frames a Pokemon’s information has been displayed for after a classification. We’ll be comparing the value to CONFIG["display_for"]  to determine if we are finished displaying a result.

The predicted class label (Pokemon species) will be held in predLabel  — it is initialized to None  for now.

To handle the clicking/tapping on the user interface, OpenCV requires us to define a callback function. Our on_click  callback method is defined below:

The on_click  callback function is executed each time a click is captured by the GUI. We’re only concerned with the click event parameter; however, OpenCV provides additional information such as the (x, y)-coordinates, flags, and param.

When our callback function encounters a left mouse click or a finger tap ( event == cv2.EVENT_LBUTTONDOWN ), we preprocess  and classify  our frame , storing the result as predLabel  and marking clicked  as True . As denoted by Line 61, frame , clicked , and predLabel  are global variables.

The preprocess  steps must be identical to the steps taken when training our model (you can learn how we trained our Keras deep learning model in an earlier post in the series).

Below you can find the preprocess  method:

In this method, our first step is to  resize  the frame/ image  to (96, 96)  pixels. Since our model is designed for these dimensions, we must use the same dimensions.

Next, we scale the image pixel array values to the range of [0, 1] .

Subsequently, we call img_to_array  on the image which orders the channels of the array properly, based on “channels first” or “channels last” ordering.

We train/classify images in batches. After calling np.expand_dims  on the image, it will have the shape (1, 96, 96, 3) . Forgetting to add in this extra dimension will result in an error when calling our predict  method of the model in the classify  function.

Lastly, we return the image  to the calling function.

Note: Do the the preprocessing steps of the preprocess  function look foreign to you? Preprocessing is essential to all deep learning workflows. I cover preprocessing in depth inside of Deep Learning for Computer Vision with Python (along with many more topics that will take you from beginner to seasoned expert throughout the course of the book bundles). Be sure to check out the free sample chapters available here.

Our final helper function, classify , accepts an input image  (making the assumption that it has already been pre-processed) and then classifies it:

Our classify function is very direct and to the point, but this is actually where all the magic happens under the hood.

Calling model.predict  on the image and grabbing the zero-index result (the results for the first and only image in the batch) returns a list of each of probabilities from the softmax layer in our network (Line 81).

Taking the index of the maximum probability and feeding it into our labels list (in CONFIG ) yields the human readable class label (Line 84). We then return this label to the calling function.

Now that our helper functions are defined we can move on to creating the user interface:

We first load our user interface image from disk and extract the height and width (Lines 88 and 89).

From there we load the mask image from disk (Line 94) and convert it to a single channel grayscale image (Line 95).

We then apply a binary threshold (Lines 96 and 97). After thresholding, the image will only contain 0’s and 255’s (0 for black background and 255 for white foreground).

Moving on, let’s load data, initialize objects, and setup our callback function:

We load the Pokedex CNN model on Line 100 and the Pokemon database on Line 101.

Then, we initiate our VideoStream  object. I’m using the PiCamera as is shown on Line 106. If you’re running the app on your laptop/desktop, you can comment this line out (Line 106) and uncomment Line 105.

We pause for 2.0  seconds to allow for the camera to warm up (Line 107).

From there, we need to setup the mouse callback listener function. In order to do accomplish this, we first need a namedWindow . I named the window "Pokedex"  on Line 110 and then established the mouse callback on Line 111.

Let’s begin processing frames in a while  loop:

Inside of the while True  loop, we first check to see if we are currently displaying a classification (Lines 117-118) and if so, increment the count  variable. In other words, if this if statement is triggered, the frame won’t change up until the number of CONFIG["display_for"]  frames.

Otherwise, let’s grab a new frame  from the video stream process it. First we resize it and extract the dimensions (Lines 124 and 125). We also reset count , clicked , and predLabel  (Lines 129-131).

Going back to the main execution flow of the loop, we create the actual user interface from the frame:

First, we allocate a  pokedex  array with same dimensions as the background image (Line 136).

Then, we store the frame  in the pokedex  array using the coordinates specified from our configuration dictionary on Lines 137 and 138. Essentially, this puts the frame where the white box resides in Figure 9.

We now have a masking trick to perform. The goal of the next two lines is to achieve rounded corners like the white box in Figure 9.

To accomplish the rounding of corners, we first compute the bitwise_and  between the pokedex  image and the pokedexMask  (Line 144). This produces round corners and removes any frame content that falls outside the viewport of the mask.

Then, the bitwise_or  is taken to combine both the pokedex  and pokedexBG  to form the final user interface image (Line 145). This bitwise_or  only works because the pokedexBG  has a value of 0 (black) for the screen viewport region.

Note: To learn more about masking, as well as other image processing fundamentals, refer to Practical Python and OpenCV + Case Studies and/or the PyImageSearch Gurus course.

If a classification has just taken place, we need to draw the Pokemon name, height, and weight on the UI:

So long as the predLabel  is populated with a class label, we’re going to draw the class label text and lookup other relevant information in our Pokemon database JSON file to display.

Lines 151-153 handle drawing the Pokemon species text (also known as the CNN class label).

Similarly Lines 156-158 and Lines 161-163 handle drawing the Pokemon’s height and width respectively.

Let’s show the output frame and perform cleanup:

On Line 166 we display the result of our hard work in the "Pokedex"  named window.

We grab keypresses (Line 167) and if the "q"  key has been pressed, we break  out of the loop (Lines 170 and 171).

Upon exiting the while loop, we perform cleanup (Lines 175 and 176) and exit the script.

Real-time deep learning results

We are now ready to see our real-time deep learning application on the Raspberry Pi in action!

Make sure you use the “Downloads” section of this blog post to download the source code.

From there, open up a shell and execute the following command:

If you’re using a Raspberry Pi to run this code it may take 20-30 seconds to initialize TensorFlow and import the Keras model itself.

Be patient!

Once the model is loaded into memory you’ll be able to perform deep learning image classification in real-time.

A full demo of the Pokedex + real-time deep learning model in action can be found below:

I also created a simple Python script (that ran in the background) to take a screenshot every two seconds — I pieced the screenshots together to form an animated GIF of classifying Charmander:

Figure 10: Our deep learning Pokedex project correctly recognizes Charmander.

In my Pokemon collection, I have a 3D printed Bulbasaur that I decided to classify as well:

Figure 11: Our Keras CNN is capable of recognizing a 3D printed Bulbasuar!


In today’s post we:

  1. Learned how to perform real-time deep learning on the Raspberry Pi
  2. Implemented a fun, hands-on deep learning project

All of the hardware and components, including the Raspberry Pi itself, are totally optional. You can just as easily execute this code on your laptop or desktop provided you have either a built-in webcam or USB camera.

I chose to use the Raspberry Pi for this project as:

  1. I receive a number of questions from readers asking if real-time deep learning is possible on the Raspberry Pi (it is, provided your model is simplistic enough).
  2. I wanted to build an actual device rather than just an application.

One of my favorite aspects of running PyImageSearch is being able to use my imagination, incorporate a bit of nostalgia, and build actual fictional applications and devices from my childhood.

These devices, such as the Pokedex, were once thought make believe — but by using deep learning, we’ve made them a reality.

I hope you enjoyed this series of blog posts and thank you for taking this journey with me.

But before you go, make sure you enter your email address in the form below to join the PyImageSearch newsletter. I’ll send you a quick email each week new posts are published here on PyImageSearch.


If you would like to download the code and images used in this post, please enter your email address in the form below. Not only will you get a .zip of the code, I’ll also send you a FREE 17-page Resource Guide on Computer Vision, OpenCV, and Deep Learning. Inside you'll find my hand-picked tutorials, books, courses, and libraries to help you master CV and DL! Sound good? If so, enter your email address and I’ll send you the code immediately!

, , , , , , ,

26 Responses to A fun, hands-on deep learning project for beginners, students, and hobbyists

  1. Tim April 30, 2018 at 11:09 am #

    Adrian, I love this series, I hope to use the process to train for other recognition (maybe detect which family member is coming through the door?)

    Would adding the NCS improve performance here? (can the Keras model even be used on the NCS?)

    • Adrian Rosebrock April 30, 2018 at 11:42 am #

      Keras models cannot be directly used on the NCS. You would first have to take the weights and convert them to a TensorFlow-specific format so they can be ran on the NCS. I’ve heard Intel and Movidius will try to support Keras models directly in a future release.

      Our model is small enough that the NCS wouldn’t be a huge help but that’s also how the app is created. We tap the screen, capture a single image, and pass it through the network. The actual inference time is extremely low and it could even run in ~10 FPS without an NCS.

    • David Hoffman May 1, 2018 at 1:35 pm #

      Hi Tim, I’ve heard of success with this GitHub project for converting Keras models to graph files, but I haven’t tried it myself.

  2. Anirban Ghosh April 30, 2018 at 12:26 pm #

    great post

    • Adrian Rosebrock April 30, 2018 at 12:53 pm #

      Thanks Anirban!

  3. Oleg Novikov April 30, 2018 at 2:11 pm #

    Adrian, thank you so much for this tutorials! They are great!
    I also bought NCS too.
    Do you have any suggestion on which way to go with more or less accurate real time facial recognition?
    Will CNN and Karas still fit with that type of project or should I try NCS and something else?
    Thank you so much again!
    Your tutorials is something I was really looking for!

    • Adrian Rosebrock April 30, 2018 at 6:34 pm #

      1. Take a look at facial recognition algorithms such as Eigenfaces, Fisherfaces, LBPs for facial recognition, and deep learning embeddings. I cover these methods in detail, including real-time implementations, inside the PyImageSearch Gurus course.

      2. Keras models cannot (currently) be directly used on the Intel NCS, as I mentioned in a previous comment on this post. See my reply to Tim for more details.

  4. Raymond KUDJIE April 30, 2018 at 4:29 pm #

    Awesome tutorial!!!!

    • Adrian Rosebrock April 30, 2018 at 6:31 pm #

      Thanks Raymond! 🙂

  5. issaiass April 30, 2018 at 6:11 pm #

    Nice project Adrian. UX/UI could be improved if you had used QtCreator too, it gives you a very nice look n feel.

    • Adrian Rosebrock April 30, 2018 at 6:32 pm #

      To be honest I don’t have a lot of experience with GUI libraries, but yes, using a dedicate GUI library would certainly give the project a better look/feel.

      • issaiass April 30, 2018 at 7:15 pm #

        You should try Qt Creator maybe once, if not… its free for non-commercial use.

        Not only for the look n feel, its all the libraries, tools, data manipulation, web, multimedia manipulation, bindings, documentation, examples, etc… and you off course could use python or pyqt for your development

        • Adrian Rosebrock May 1, 2018 at 8:30 am #

          Thanks for suggestion — I may give it a try in the future 🙂

  6. Basha May 3, 2018 at 2:44 am #

    detecting mobile phone in real time is it possible

  7. Jose Luis V. May 3, 2018 at 6:27 am #

    Thank you very much, Adrian

    • Adrian Rosebrock May 3, 2018 at 9:28 am #

      Thanks Jose, I’m glad you enjoyed it!

  8. Matthew teow May 8, 2018 at 10:06 pm #

    Hi Adrian, great post! Love it!

    • Adrian Rosebrock May 9, 2018 at 9:32 am #

      Thanks Matthew!

  9. Yash Reddy May 15, 2018 at 10:58 pm #

    Hi Adrian. I really love your project and I’ve decided to take it up. I have a few questions though:

    How do I run your code on Windows 10?

    If I want to add more Pokemon, how would I go about doing so?

    Would it be possible to add audio to the entries?

    I want to try and run the program on my laptop before I integrate it onto my Raspberry Pi (so I can work on the program at school). Could you explain the process a little bit more in depth when trying to run your program on Windows 10?

    Thank you for putting so much time toward trying to teach us; I definitely appreciate your expertise.

    • Adrian Rosebrock May 17, 2018 at 6:59 am #

      1. This code will run on Windows 10. Use the “Downloads” section of this blog post to download the code, unzip the archive, open a command line, navigate to where you downloaded the code, and execute the script(s).

      2. See this blog post where we trained the original model. It discusses the directory structure required for your dataset, including adding new classes.

      3. See #1

      • Yash Reddy May 18, 2018 at 4:17 pm #

        When I tried running straight from Windows Command Prompt, I got an error: “File “C:\Users\yvall\Documents\pokedex\pokedex.py”, line 5, in
        from keras.preprocessing.image import img_to_array
        ImportError: No module named keras.preprocessing.image”
        Are there other programs I need to download in order to run this script? I want to try your Pokedex program on my Windows laptop that has a built-in webcam before I eventually run this program on the Raspberry Pi.

        While going through the rest of your tutorials, would I be fine if I run Ubuntu 16.04 and Python 2.7 on VMware, non-commercial use?

        Thank you Adrian.

        • Adrian Rosebrock May 22, 2018 at 6:30 am #

          It sounds like you may not have Keras installed on your system. Make sure you install Keras before proceeding.

          To answer your second question, yes, you can use a VM.

  10. Tim October 21, 2018 at 10:12 am #

    Morning Adrian, great tutorials! – I was trying a twist of this with only two classes of images in the dataset, but it comes back with an error of …line 138, in standardize_input_data
    ValueError: Error when checking target: expected activation_7 to have shape (2,) but got array with shape (1,)

    – I’ve burned a good bit of time trying to chase down exactly what this is trying to tell me and how to resolve it, but haven’t had any success yet. Would you mind pointing me in the right direction? – I went back to your original source code/example set and tried doing the same thing, and with only two pokemon folders I run into the same problem.

    Thanks in advance for any help you can provide, and for the guides!

    • Adrian Rosebrock October 22, 2018 at 8:00 am #

      Hey Tim, go back to your code and check how you are parsing the labels from the input image paths. I’m 99% sure there is a bug in your code where you are incorrectly parsing the labels (i.e., the code thinks there is only 1 class label but you’re telling Keras there are 2).

Before you leave a comment...

Hey, Adrian here, author of the PyImageSearch blog. I'd love to hear from you, but before you submit a comment, please follow these guidelines:

  1. If you have a question, read the comments first. You should also search this page (i.e., ctrl + f) for keywords related to your question. It's likely that I have already addressed your question in the comments.
  2. If you are copying and pasting code/terminal output, please don't. Reviewing another programmers’ code is a very time consuming and tedious task, and due to the volume of emails and contact requests I receive, I simply cannot do it.
  3. Be respectful of the space. I put a lot of my own personal time into creating these free weekly tutorials. On average, each tutorial takes me 15-20 hours to put together. I love offering these guides to you and I take pride in the content I create. Therefore, I will not approve comments that include large code blocks/terminal output as it destroys the formatting of the page. Kindly be respectful of this space.
  4. Be patient. I receive 200+ comments and emails per day. Due to spam, and my desire to personally answer as many questions as I can, I hand moderate all new comments (typically once per week). I try to answer as many questions as I can, but I'm only one person. Please don't be offended if I cannot get to your question
  5. Do you need priority support? Consider purchasing one of my books and courses. I place customer questions and emails in a separate, special priority queue and answer them first. If you are a customer of mine you will receive a guaranteed response from me. If there's any time left over, I focus on the community at large and attempt to answer as many of those questions as I possibly can.

Thank you for keeping these guidelines in mind before submitting your comment.

Leave a Reply