Project Idea
For my project, I chose to write code for a model that generates images of trees based on photographs I took. The choice of this specific solution is quite straightforward: for many of us, AI is associated with a soulless machine incapable of perceiving the beauty of nature or recreating it on its own. Therefore, I decided that I wanted to create a kind of meaningful image series this way.
While writing the code, I utilized both the initial notebook from the course and the preliminary version of Module 4's project; a link to this will be included in the sources list at the end. I also received assistance from other neural network models, specifically Google AI, ClaudAI, and Gigachat. To ensure the code functioned correctly, I prompted the AI to help debug errors in the programs and suggest ways to make the model training process faster and more efficient.
Program Development Process Description
In this section, we will attach explanations detailing the purpose of each command within the program.
These screenshots illustrate the first two steps in model training—specifically, verifying whether the processor intended for the AI is powerful enough, as well as preparing the working environment for the subsequent training and image generation processes.
Below are screenshots of the commands needed to run and load the dataset itself. You can also see some of the images I uploaded to the model for subsequent training (one of the photos may differ, as the file was lost during the process and had to be replaced with another photo).
In this part of the program, you can see the beginning of training a model based on uploaded photographs: the code’s goal is first to process the photos, then to load textual descriptions for them, and finally to move on to generating new images.
And finally, the last part of the program: the screenshot only shows the section for generating the first sample image, but in the code file itself, you can see the part of the model designed to generate a series of images intended for the project.
Generation Examples
The first generated example
So, this is the result of training the model on tree photographs. I did not use any additional image enhancement methods, so the result remains as it is. Although the images clearly show that they were generated by a neural network, they possess a unique atmosphere, a quite realistic composition, and generally a resemblance to real photographs. Based on this, I can conclude that the objective set at the beginning of the model development was achieved: in the generations compiled in this project, you can see echoes of real natural beauty—beauty that seemed inaccessible to artificial intelligence.
Sources
- Link to last module code:
https://drive.google.com/drive/folders/1GYmgKhZ6BPJgQKuQa5mQodaFPwKdP85m?usp=drive_link
-GigaChat:
https://giga.chat
-GoogleAI (built into Chrome)
-ClaudeAI:
https://claude.ai
