Reading Time

20 Minutes

Abduragim Shtanchaev

Developer at OpenCV.ai

Debugging OpenCV with ChatGPT: A Developer's Guide

Getting the Hang of OpenCV’s Inner Workings with ChatGPT

Keeping pace with the evolution of technology and utilizing it judiciously, this blog explores how ChatGPT can serve for code development debugging.

April 10, 2024

Introduction

As programmers, we often work with familiar development environments, but occasionally we encounter new tools that can be time-consuming and challenging to learn. In such situations, having virtual assistance can be extremely beneficial. Integrating opencv chatgpt into our workflow brings that much-needed support, seamlessly fitting into our toolkit to ease the learning curve.

In this article, I will share my experience of contributing to OpenCV, a renowned open-source library, despite having limited knowledge of C++ and understanding its architecture. I achieved this with the assistance of ChatGPT, a Large Language Model (LLM).

Building OpenCV

Let’s begin with a simple task - building OpenCV. While there are existing resources for this, I’ll demonstrate how ChatGPT can generate an installation guide for OpenCV for any platform (Windows, Linux, or Mac) by adjusting the prompt. The answers will be included in the expanding section.

Note: Some answers from the free version of ChatGPT3.5 might be misleading and will be marked with an exclamation mark: ❗ accompanied by corrections.

Prompt:

How to build OpenCV locally on Ubuntu, you can follow these steps.

❗Click to see the GPT answer:

Update your system: Open a terminal and run the following commands to update your Ubuntu system:
```
        sudo apt update
```

Install dependencies: Install the necessary dependencies for building OpenCV by running the following command:

        sudo apt install build-essential cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev

Clone the OpenCV repository: Use Git to clone the OpenCV repository to your local machine:
```
        git clone https://github.com/opencv/opencv.git
```
Create a build directory: Navigate to the cloned OpenCV repository and create a build directory:
```
        
  cd opencv
  cd build
        
```
Configure the build: Run CMake to configure the build settings. CMake will generate the necessary build files based on your system configuration:
```
        cmake ..
```
You can customize the build by specifying additional options to the cmake command. For example, to enable Python support, you can use:
```
        cmake -D BUILD_opencv_python2=ON -D BUILD_opencv_python3=ON ..
```
OpenCV maintainer’s note: Bindings for Python 2.x are deprecated and not recommended for new projects. OpenCV requires Numpy and Python development files. You need to install python3-dev and python3-numpy for a successful build. Also, the OpenCV team develops an OpenCV-Python package for pip. Please follow the project instructions to build a pip-manageable installation from the source.
Build OpenCV: Once the configuration is complete, you can start the build process:
```
        make -j$(nproc)
```
The -j$(nproc) flag enables parallel compilation using all available CPU cores, which can significantly speed up the build process.
Install OpenCV: After the build is complete, you can install OpenCV on your system:
```
        sudo make install
```
Configure dynamic linker: To ensure that the OpenCV libraries are correctly linked, run the following command to configure the dynamic linker:
```
        sudo ldconfig
```
Verify the installation: You can verify the installation by compiling and running a sample OpenCV program. For example, create a file named test.cpp with the following content:
```
        #include <opencv2/opencv.hpp>

    int main() {
        // std::string image_path = "path-to-image"
        // cv::Mat image = imread(image_path, cv::IMREAD_COLOR)

        cv::imshow("Test", image);

        return 0;
  }
```
Compile the program using the following command:
```
        g++ test.cpp -o test `pkg-config --cflags --libs opencv`
```
OpenCV maintainer’s note: Modern OpenCV (4.x) does not generate a package config file by default. The proposed solution does not work. The recommended way is using CMake.

Replace "path/to/your/image.jpg" with the path to an actual image file on your system. Execute the compiled program:
```
        ./test
```
If the program displays the image correctly, it means that OpenCV is successfully installed and configured on your Ubuntu system.

By following these steps, you can build OpenCV locally on Ubuntu and start using it for your computer vision projects.

However, it’s important to acknowledge that using the ChatGPT may sometimes lead to incorrect results. If you are unfamiliar with the expected answers, navigating with the model’s responses can be challenging. For instance, the provided CMake flags are not what is actually expected. Let’s clarify the correct CMake flags below:

Prompt:

How should cmake flags be set normally?

❗Click to see the GPT answer:

When configuring the build of OpenCV using CMake, you can set various flags to customize the build process according to your requirements. The choice of flags depends on your specific use case and the features you want to enable or disable in OpenCV. Here are some commonly used CMake flags for OpenCV:

General build options:

-DCMAKE_BUILD_TYPE: Specifies the build type, such as Debug, Release, or RelWithDebInfo

-DCMAKE_INSTALL_PREFIX: Sets the installation directory for OpenCV

-DBUILD_SHARED_LIBS: Builds OpenCV as shared libraries (ON) or static libraries (OFF)
Language bindings:

-DBUILD_opencv_python2: Enables Python 2.x support.

-DBUILD_opencv_python3: Enables Python 3.x support.

-DINSTALL_PYTHON_EXAMPLES: Installs Python examples.
Module-specific options:

-DWITH_MODULE_NAME: Enables a specific OpenCV module, such as -DWITH_CUDA for CUDA support or -DWITH_OPENGL for OpenGL support. Refer to the OpenCV documentation for a list of available modules.
❗Build optimizations and performance:

-DENABLE_FAST_MATH: Enables faster but less accurate math optimizations

-DENABLE_AVX*: Enables AVX instruction set optimization

-DENABLE_SSE*: Enables SSE instruction set optimization. ## Legacy flag. See here

-DENABLE_NEON*: Enables NEON instruction set optimization (for ARM-based systems)
❗External dependencies:

-DWITH_X**: Enables support for X Window System. ## No such in OpenCV

-DWITH_TBB: Enables support for Intel Threading Building Blocks.

-DWITH_OPENMP: Enables support for OpenMP parallelism.

-DWITH_QT: Enables support for Qt GUI framework.

-DWITH_OPENGL: Enables support for OpenGL.

These are just a few examples of commonly used CMake flags. You can find a more comprehensive list of CMake flags specific to OpenCV in the OpenCV documentation.

When configuring your build, it's essential to strike a balance between enabling the necessary features for your project and optimizing the build for your system. Enabling unnecessary features may increase the build time and result in larger binary sizes, so it's advisable to only enable the features you need.

Experiment with different combinations of CMake flags based on your project requirements, hardware capabilities, and the availability of external dependencies to obtain the desired configuration for your OpenCV build.

OpenCV maintainer’s note:

ENABLE_XXX options for hardware-specific optimization branches exist in OpenCV 4.x, but are marked as deprecated. OpenCV uses a more flexible mechanism for static optimizations and run-time dispatch. All options are described on the project wiki.
There is no WITH_X option in OpenCV CMake. It’s a ChatGPT mistake.

The model’s responses are more informative and helpful, providing a starting point for the users to build the library for the first time. Asking specific questions allows getting tailored instructions, which can save a lot of time compared to general web searches.

Understanding OpenCV’s Structure

When dealing with new libraries, it’s essential to understand their structure and modules. ChatGPT can also be helpful in this aspect. Let’s explore how it handles OpenCV:

Prompt:

Could you provide an overall structure of the OpenCV library's repository?

❗Click to see the GPT answer:

The OpenCV library's repository follows a modular structure, with each module residing in a separate directory. Here is an overview of the overall structure of the OpenCV repository:

Modules: The repository contains a directory for each module, where the main functionality of OpenCV is organized. Some of the core modules include:

- opencv_core: Core functionality and basic data structures

- opencv_imgproc: Image processing functions

- opencv_video: Video analysis and processing

- opencv_highgui: User interface and image/video I/O

- opencv_calib3d: Camera calibration and 3D reconstruction

- opencv_features2d: Feature detection and description algorithms

- opencv_objdetect: Object detection and recognition

- opencv_ml: Machine learning algorithms and tools

- opencv_dnn: Deep Neural Networks module

- opencv_photo: Computational photography algorithms

- opencv_stitching: Image stitching and panorama creation

- opencv_videoio: Video I/O functionality

In addition to these core modules, several additional modules provide additional functionalities like bio-inspired vision, 3D object recognition, and more.

Additional directories:

- 3rdparty: Third-party dependencies used by OpenCV

- apps: Applications and examples showcasing the usage of OpenCV

- cmake: CMake scripts for building OpenCV

- data: Additional data files used by OpenCV, such as Haar cascades

- doc: Documentation, including tutorials, guides, and API references

- modules: CMake build scripts for each module

- platforms: Platform-specific files

- samples: Code samples and examples

- test*: Unit tests for OpenCV
Build files:

- CMakeLists.txt: The main CMake build file that configures the build process for the entire repository.

- CMakeLists.txt.in*: Template CMake build file used for generating the actual `CMakeLists.txt` during the build configuration.

OpenCV maintainer’s note: All OpenCV CMakeLists.txt files are stored in git. OpenCV does not generate CMakeLists.txt from templates. All tests are placed in a module-specific folder. There is no global ‘test’ folder.

The modular structure of the OpenCV repository allows for easier maintenance, independent development of each module, and flexibility in building and using only the required functionality.

It's important to note that the repository structure may evolve as new modules and functionalities are added or modified. It's always recommended to refer to the official OpenCV repository for the most up-to-date information on the repository structure and organization.

The overall structure of the OpenCV library is correctly explained, providing a good starting point for navigation.

Now, let’s test something more challenging and see how ChatGPT performs or fails.

Prompt:

How do run tests written particularly for the dnn module?

❗Click to see the GPT answer:

To run tests specifically written for the DNN (Deep Neural Networks) module in OpenCV, you can follow these steps:

Build OpenCV: Make sure you have built OpenCV with the DNN module enabled. Follow the instructions for building OpenCV locally, ensuring that you include the DNN module during the configuration step.
Locate the test files*: In the OpenCV source code, navigate to the /modules/dnn/test directory. This directory contains the test files specifically written for the DNN module.

Build the tests*: Use CMake to generate the build files for the tests. Create a build directory and run CMake with the path to the OpenCV source code and the build directory:

            
 cd opencv/build  # Navigate to your OpenCV build directory
 cmake /path/to/opencv/source  # Replace /path/to/opencv/source with the actual path to the OpenCV source code

OpenCV maintainer’s note: All accuracy tests are located in the module folder and the DNN test code location is correct, but they could not be built in the proposed way. OpenCV builds all tests during the common build procedure by default. So step #3 is not needed.

Compile the tests: Once the configuration is complete, build the tests using the make command:
```
             make -j$(nproc)
```

This command compiles the tests in parallel using all available CPU cores.

Run the tests*: After the tests are built, you can execute them using the `make test` command:
```
            make test
```

This command will run all the tests for the DNN module.

OpenCV maintainer’s note: OpenCV testing requires test data stored in a separate repository opencv_extra and a large set of public DNN models for solid test coverage. Please follow this instruction on the project wiki.

Analyze the test results: Once the tests are complete, you can analyze the test results to check for any failures or errors. The test framework will provide an output indicating the success or failure of each test case.

By following these steps, you can build and run the tests specifically written for the DNN module in OpenCV. This helps ensure the correctness and functionality of the DNN module in your OpenCV installation.

The provided answer may appear reasonable at first. However, upon testing, it becomes evident that it does not work as expected.

Tests for OpenCV are generated during build and are located in /build/bin/test_

Moreover, the response lacks information about the test data used in these tests. This highlights the importance of being cautious with the answers provided by LLMs, as they can sometimes produce false or incomplete information.

Debugging OpenCV LLM

Let’s focus on a specific problem encountered during my OpenCV contribution. The solution cannot be found directly via Google, so let’s explore how ChatGPT helped me debug it.
I had an issue with parsing the ONNX graph with OpenCV for an LSTM model (onnx graph of LSTM model that has been trained using PyTorch) while making the model inference. Below is a Python script to showcase the problem:


import torch
import torch.nn as nn
import onnxruntime
import numpy as np
import cv2
import random

class LayerLSTM(nn.Module):
    def __init__(self, features, hidden_size, layers=1, batch_first=False):
        super(LayerLSTM, self).__init__()
        self.rnns = nn.LSTM(
            input_size=features,
            hidden_size=hidden_size,
            num_layers=layers,
            batch_first=batch_first
        )
        self.set_weights_to_ones()

    def forward(self, x, hx, cx):
        x, (hx, cx) = self.rnns(x, (hx,cx))
        return x, (hx, cx)


if __name__ == "__main__":

    # Set random seeds
    seed = 42
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.benchmark = False
    np.set_printoptions(precision=15)

    features    = 3
    hidden_size = 7
    batch_size  = 5
    seq_len     = 2
    layout      = True

    model = LayerLSTM(features, hidden_size, batch_first=layout)

    x  = torch.ones(seq_len, batch_size, features)
    hx = torch.ones(1, batch_size, hidden_size)
    cx = torch.ones(1, batch_size, hidden_size)

torch.onnx.export(
		model,
		(x, hx, cx),
		'./sample_lstm.onnx',
		verbose=True,
		input_names=['x', 'hx', 'cx'],
		output_names=['output'])

net = cv2.dnn.readNetFromONNX("./lstm_model.onnx") # ------ fails here!!!

with error:


[ERROR:0@0.087] global onnx_importer.cpp:1056 handleNode DNN/ONNX:
ERROR during processing node with 7 inputs and 3 outputs: 
[LSTM]:(onnx_node!/rnns/LSTM) from domain='ai.onnx'

Traceback (most recent call last):
File "test_lstm.py", line 40, in 
net = cv2.dnn.readNetFromONNX('./sample_lstm.onnx')
cv2.error: OpenCV(4.7.0-dev)

/opencv/modules/dnn/src/onnx/onnx_importer.cpp:1075: error: (-2:Unspecified error) in function 'handleNode'
Node [LSTM@ai.onnx]:(onnx_node!/rnns/LSTM) parse error: OpenCV(4.7.0-dev)
/opencv/modules/dnn/src/onnx/onnx_importer.cpp:593: error: (-5:Bad argument) Blob hx not found in const blobs in function 'getBlob'

The error log shows that the Blob hx could not be found in the const blobs within the getBlob function. I used a debugger to trace back the issue, but was not familiar with how things should work for parsing LSTM, so I reached out to ChatGPT for help.

Prompt:

Here is a log error that I get when I try to parse a simple onnx model with a single LSTM layer

    
    [ERROR:0@0.087] global onnx_importer.cpp:1056 handleNode DNN/ONNX:
    ERROR during processing node with 7 inputs and 3 outputs:
    [LSTM]:(onnx_node!/rnns/LSTM) from domain='ai.onnx'

    Traceback (most recent call last):
    File "test_lstm.py", line 40, in 
    net = cv2.dnn.readNetFromONNX('./sample_lstm.onnx')
    cv2.error: OpenCV(4.7.0-dev)

    /opencv/modules/dnn/src/onnx/onnx_importer.cpp:1075: error: (-2:Unspecified error) in function 'handleNode'
    Node [LSTM@ai.onnx]:(onnx_node!/rnns/LSTM) parse error: OpenCV(4.7.0-dev)
    /opencv/modules/dnn/src/onnx/onnx_importer.cpp:593: error: (-5:Bad argument) Blob hx not found in const blobs in function 'getBlob'

What could be a possible problem here?

❗Click to see the GPT answer:

Based on the error message you provided, it appears that the ONNX model you're trying to parse has an issue with the LSTM layer. Specifically, the error seems to be related to the hx input, which is not found in the constant blobs

Here are a few possible reasons for this issue:

The ONNX model might be incorrectly exported or not supported by the OpenCV version you're using. To verify this, you can try to run the model using another ONNX-compatible framework, like ONNX Runtime or PyTorch, and check if the issue persists.
The ONNX model might use an LSTM layer with non-standard inputs, or the LSTM layer in OpenCV does not support certain configurations. You can check the model's structure using an ONNX visualization tool like Netron to see if there are any discrepancies between the ONNX model and the expected OpenCV LSTM layer format.
The OpenCV version you are using (4.7.0-dev) might have a bug or incomplete implementation of the LSTM layer. You could try using a different version of OpenCV or build the library from the latest source code to see if the issue persists.

To further diagnose and resolve the issue, you can try the following steps:

Update OpenCV: Make sure you are using the latest version of OpenCV, as it might have bug fixes and improvements related to ONNX model parsing and LSTM layers.
Verify the ONNX model: Use an ONNX model validator or ONNX Runtime to check the model's correctness and ensure that it adheres to the ONNX specifications.
Inspect the model's structure: Use an ONNX visualization tool like Netron to inspect the model's structure, especially the LSTM layer's inputs, and configuration, and compare it with the expected format in OpenCV.
Debug the ONNX Importer: Add debug logs or breakpoints in the ONNX Importer code modules/dnn/src/onnx/onnx_importer.cpp to gain more insights into the parsing process and identify the root cause of the issue.
Consult the OpenCV community: If you still can't resolve the issue, consider reaching out to the OpenCV community for assistance through forums or by reporting the issue on the OpenCV GitHub repository

‍

ChatGPT's answer was informative and provided some new ideas. To further investigate, I visualized the ONNX graph using netron.app, which clarified that hx is an input, not a constant blob.

Understanding ONNX Parsing in ChatGPT for OpenCV

A picture of the ONNX graph makes it clear that hx is the input. However, the error log indicates that hx is being searched in constant blobs. Since I knew the ONNX file was valid (it could be read by other frameworks), I knew that problem was specifically related to parsing. However, since I was unfamiliar with parsing in OpenCV, I asked ChatGPT to clarify it for me.

Prompt:

How does opencv parse the onnx graph? I want to know what happens under the hood when I call this function net = cv2.dnn.readNetFromONNX('some_path.onnx'). Please, give me an explanation of which modules and which classes are being called.

❗Click to see the GPT answer:

When you call net = cv2.dnn.readNetFromONNX('some_path.onnx'), the OpenCV DNN (Deep Neural Network) module parses the ONNX (Open Neural Network Exchange) model and constructs a computational graph. Under the hood, several classes and functions are called to achieve this. Here's an overview of the process:

First, the function in the cv2.dnn module is called. This function acts as a high-level wrapper that delegates the task of reading and parsing the ONNX file to the internal OpenCV classes.
The readNetFromONNX function calls the readNet function, which initializes a Net object. This object is used to store the entire neural network graph, layers, and weights.
The readNet function then calls the ONNXImporter class, which is responsible for parsing the ONNX file and extracting the necessary information to construct the computational graph.
The ONNXImporter class reads the ONNX file and converts it into an internal OpenCV representation using the protobuf library. The ONNX file contains the model architecture and weights in the form of an ONNX graph.
The ONNXImporter class iterates through the nodes of the ONNX graph and converts them into OpenCV DNN layers. It does this by mapping the ONNX operators to the corresponding OpenCV DNN layer types. For each node, the importer extracts the layer's parameters and initializes an OpenCV DNN layer with those parameters.
The ONNXImporter class then adds the created OpenCV DNN layers to the Net object, connecting them according to the graph structure defined in the ONNX file.
Once all the layers are added and connected, the Net object is returned to the caller. This object can now be used to perform inference using the forward method

We understand how the ONNX parser works (steps 5-6 in ChatGPT's answer), but I still need help with the functions inside the ONNXImporter class in OpenCV. Can ChatGPT clarify any unclear code? I added the beginning of the ONNXImporter class code for reference to the prompt:

Prompt:

Can you explain this class to me? Specifically which function is responsible for parsing ONNX graph?

    
    class ONNXImporter
    {
    FPDenormalsIgnoreHintScope fp_denormals_ignore_scope;

    opencv_onnx::ModelProto model_proto;
    struct LayerInfo {
        int layerId;
        int outputId;
        int depth;
        LayerInfo(int _layerId = 0, int _outputId = 0, int _depth = CV_32F)
            :layerId(_layerId), outputId(_outputId), depth(_depth) {}
    };


    ...
    ...
    ...

What could be a possible problem here?

❗Click to see the GPT answer:

This is a class called ONNXImporter that is responsible for importing a neural network model in the ONNX format and converting it to an OpenCV neural network

The main function responsible for parsing the ONNX graph is the handleNode() function. It takes a NodeProto object that represents a single node in the ONNX graph and parses it according to its operation type, creating a corresponding layer in the OpenCV neural network.

The ONNXImporter class also contains several utility functions such as getBlob(), which retrieves a blob (a tensor) from the constant blobs map based on its name, and getLayerParams(), which parses a NodeProto object and returns a LayerParams object containing the layer parameters.

Other functions in the class handle various types of ONNX operators, such as parseConv(), which handles convolution layers, and parseLSTM(), which handles LSTM layers.

The class also contains several private data members, such as the model_proto object, which stores the ONNX model, and the constBlobs map, which stores the constant blobs of the model.

It's clear that the handleNode() function in the ONNXImporter extracts weights from each node using getBlob().

Upon reviewing the error log once again, it became apparent that the LSTM parser expected hx to be inside the const blobs weight, but it is an input, as we saw in the ONNX graph. Therefore, the bug exists inside the LSTM parser function! I also checked ONNX documentation (see initial_h) to verify that hx indeed can be passed as input to the layer.

    
    /opencv/modules/dnn/src/onnx/onnx_importer.cpp:593:
    error: (-5:Bad argument) Blob hx not found in const blobs in function 'getBlob'

I had difficulty finding where the getBlob function was called inside the handleNode function. Let's see how ChatGPT clarified this:

Prompt:

Here is handleNode function, I do not see where getBlob function is being called.

    
    void ONNXImporter::handleNode(const opencv_onnx::NodeProto& node_proto)
    {
        std::cout << "inside handle node function" << std::endl;
        CV_Assert(node_proto.output_size() >= 1);
        const std::string& name = extractNodeName(node_proto);
        const std::string& layer_type = node_proto.op_type();
        const std::string& layer_type_domain = getLayerTypeDomain(node_proto);
        const auto& dispatch = getDispatchMap(node_proto);

    ...
    ...
    ...

❗Click to see the GPT answer:

It seems that the getBlob function is not called in this code snippet. However, it may be called by one of the functions being called inside handleNode that is not shown in the provided code.

Without seeing the entire implementation of the ONNXImporterclass, it is difficult to determine if and where the getBlob function is being called.

The provided answer was not helpful. Based on my own review of the code, I located where getBlob() was called:

Prompt:

I anticipate it is called here

    
    DispatchMap::const_iterator iter = dispatch.find(layer_type);
    if (iter != dispatch.end())
    {
        CALL_MEMBER_FN(*this, iter->second)(layerParams, node_proto);
    }

but not sure how this calls the function getBlob()

❗Click to see the GPT answer:

You are correct, the getBlob() function is not called directly in the handleNode() function. However, it is being called indirectly by one of the member functions of the ONNXImporter class that is being called through the dispatch map.

In the code you posted, the dispatch variable is obtained by calling the getDispatchMap(node_proto) function, which returns a map of layer types to member function pointers of the ONNXImporter class. The layer_type variable is then used to look up the appropriate member function pointer in the dispatch map.

The CALL_MEMBER_FN(*this, iter->second)(layerParams, node_proto) line calls the member function that corresponds to the layer_type key in the dispatch map. This is done using a macro CALL_MEMBER_FN which calls the member function using the function pointer stored in the iter->second

The member function may be called indirectly by handleNode() and makes use of the getBlob() function to obtain input or output blobs for the layer being processed.

‍

Things are going well so far! I found the bug and know how the LSTM layer should work. Now, I can debug the LSTM parser. ChatGPT helped me see how the parser should work in the best situation and what I should fix to make that happen. This is great! It would take me a long time to figure this out if I did it alone without ChatGPT

I propose following fix: If the hx parameter is not found in the "const blobs" (meaning it is been passed as an input), initialize it with an empty blob and assign its value to the "forward" function from an input. Otherwise, it is initialized with the values found in const blobs. That is it! I also added an accuracy tests to check the correctness of the solution. For more details here is the pull request.‍

This is how I debugged my first issue to build OpenCV models with the help of ChatGPT. It took me some time to understand how ONNXImporter and LSTM parser work with the help of ChatGPT. Eventually, I found the root cause of the issue and proposed a fix. ChatGPT helps us understand and solve problems faster. This technology can greatly free up our time.

Conclusion

This blog post explains how LLMs like ChatGPT can assist to find bugs in unfamiliar environments. OpenCV is an open-source project, so its code could have been used to train ChatGPT. However, not all projects have this benefit. For them, providing a relevant code snippet helps the model reason properly.

While LLMs like ChatGPT can help solve complex issues, they should not be the only tool for debugging. It's crucial to understand the code and the problem behind it, and test the solutions.

In summary, combining the power of LLMs with conventional techniques can be valuable in debugging unfamiliar environments. However, caution is necessary, and solutions should undergo thorough testing and verification.

Unlock the potential that AI integration offers your projects in various fields, leveraging OpenCV.ai's mastery in computer vision services. Our dedicated team is focused on harnessing AI solutions to transform and modernize practices across a wide range of industries.

Let's discuss your project

Book a complimentary consultation

Read also

April 12, 2024

Digest 19 | OpenCV AI Weekly Insights

Dive into the latest OpenCV AI Weekly Insights Digest for concise updates on computer vision and AI. Explore OpenCV's distribution for Android, iPhone LiDAR depth estimation, simplified GPT-2 model training by Andrej Karpathy, and Apple's ReALM system, promising enhanced AI interactions.

April 11, 2024

OpenCV For Android Distribution

The OpenCV.ai team, creators of the essential OpenCV library for computer vision, has launched version 4.9.0 in partnership with ARM Holdings. This update is a big step for Android developers, simplifying how OpenCV is used in Android apps and boosting performance on ARM devices.

April 4, 2024

Depth estimation Technology in Iphones

The article examines the iPhone's LiDAR technology, detailing its use in depth measurement for improved photography, augmented reality, and navigation. Through experiments, it highlights how LiDAR contributes to more engaging digital experiences by accurately mapping environments.