Hello trend,
This is Satya Mallick from LearnOpenCV.com.
In today's post, we will be build a custom detector for potholes using YOLOv7. We will train 4 models:
- yolov7-tiny using fixed resolution.
- yolov7-tiny using multiple resolution.
- yolov7 using fixed resolution.
- yolov7 using multiple resolution.
As discussed in our previous post on YOLOv7, YOLOv7 improves speed and accuracy by introducing several architectural reforms and according to the paper, it is now the new state-of-the-art in Object Detection.
In case you missed our previous posts, we have written multiple posts on Object Detection and Custom Object Detectors using YOLO:
- YOLOv7 Paper Explanation and Inference.
- Custom Object Detection using YOLOv5 on a Traffic Dataset.
- Pothole Detection using YOLOv4.
What is multi-resolution training?
In multi-resolution training, the images are be resized to +-50% of the base resolution. So, for a base size of 640×640, the minimum resolution will be 320×320 and the maximum resolution will be 1280×1280 during training. Generally, this helps to train a more robust model especially for cases when we have smaller objects. But we also need to train longer as the dataset becomes more difficult because of the varied sizes.
Without further ado, let's get started.
| YOLOv7 Custom Object Detector |
You can check out the code by clicking on the button below, and star us on GitHub to say thanks!
| Download Code (GitHub) |
AI and Creativity
AI will create many new jobs, but at the same time it will annihilate many jobs. The people it creates jobs for will be different from the people it displaces from their jobs.
We are often lulled into believing that our job is safe from an AI takeover if it involves creativity.
We don't understand the origin of creativity. Part of it appears to be inspiration derived from prior art.
But the other part appears to be truly divine not only to religious people like Ramanujan, but also to atheists like me.
That is why it felt like creative jobs will not be replaced by AI. In the last year or so, I have had to reassess that assumption.
When I got access to DALL·E 2, I was shocked to see the response to my first query "Satya Mallick in space suit."
It figured out that
- Satya Mallick is a male name.
- He is brown
But it doesn't know Satya Mallick is FAT :P.
DALL·E 2 is not the only model that helps you create art from text.
Using just a few words, you can create amazing art on on Midjourney's discord channel as done by user Lumartist#7282. See picture below.
And to top all this, Stability.AI release a public version of their diffusion model two days back.
People may not yet realize how monumental this news is for the art world. A Neural Network is challenging human creativity that builds on prior art!
AI is a transformational technology that will impact most areas of our lives. It is still in its infancy, and smart people are pouring in from different fields. That is why we see such magical growth and excitement.
If you are interested in switching your career into AI, I invite you to start with our courses. You do not need prior experience.
| JOIN THE AI REVOLUTION |
Cheers!
Satya
Courses / YouTube / Facebook / LinkedIn / Twitter / Instagram
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.