In this article, we will test Deep Block to see the power of Deep Block's patented high-resolution image processing technology.
We already explains in detail what problems we have solved.
You can check out the project used in the experiment on our platform.
Step #1 - Disappeared Parts - The Problem of Splitting
- We have already described the problems that arise in analyzing large images.
- For example, let's say a photo we need to analyze is 100,000 x 100,000 pixels.
- Many people may think that dividing the image will do the trick.
- Shall we break it into pieces?
- Even if it is not a photo like this, when we divide a high-resolution image, we must divide the objects that exist on the border of the pieces.
And, in some cases, these objects lose their characteristics. (Now the right bottom photo looks like a bird feet picture not a penguin picture)
In this case, even people cannot guess the original object by looking at the parts of the object.
Artificial intelligence models also have trouble guessing the original object by just looking at the part.
Step #2 - So what are you going to do?
- Let's start the experiment. Let's compare the difference between simply dividing images and the technology we have.
- Copy the trained machine learning model from the following link.
https://app.deepblock.net/deepblock/store/project/plilk1cyv4lteeb456
- The model has already been trained for 10 epochs.
- Now, let’s check the validation image to be fed to the trained model.
- In PREDICT mode you can check the image.
Step #3 - Simple Image Division
- Now let's start experimenting. Click the Configuration button.
- Open the modal and select the "Faster and Less Accurate" option.
This method simply divides the image.
The image is divided into 7 rows, 5 columns, and a total of 35 pieces for analysis.
Click "Divide" button.
- After division, Start the inference by clicking "PREDICT" button.
- After the inference is completed, you can see the following image.
- A few green rice paddies were not detected.
Step #4 - Accurate Analysis with Deep Block's Technology
- Let's use Deep Block's Patented Algorithm. Click the "Configuration" button.
- Open the modal and select the "Slower, More Accurate" option.
We will use the same division factor, 7 rows, 5 columns.
Click "Divide" button after choosing the option.
- After division, Start the inference by clicking "PREDICT" button.
- After the inference is completed, you can see the following image.
- It looks better than before.
Conclusion
- Our continuous dedication to advancing technology for high-resolution image analysis is evident in our work at Deep Block. Our team strives every day to develop the fastest and most accurate AI software for analyzing the world's largest images.
- Deep Block offers an innovative artificial intelligence platform tailored for large images, empowering users to effortlessly create and utilize their own machine vision models without the need for coding. Start building your own machine vision model today with ease.
DeepBlock.net