I avoid similar cat pictures and use more cat pictures for training. The model is improving. Overall, I still use too few photos. Training is done in two steps. Gray old. Red continued.
#AI #YOLO #darknet #COCO #model #training #objectdetection #boundingbox #annotation #Python #Bash #iterations
Last continued YOLO darknet training. I increased the number of training images. I removed duplicates. I also used better settings. Better mAP with less iterations.
#AI #YOLO #darknet #COCO #model #training #objectdetection #boundingbox #annotation #Python #Bash #iterations
Trainings results with continuation of training. I should have used darknet from the start.
#AI #YOLO #darknet #COCO #model #training #objectdetection #boundingbox #annotation #Python #Bash #iterations
Influence of the training iterations on the detection results.
#AI #YOLO #darknet #COCO #model #training #objectdetection #boundingbox #annotation #Python #Bash #iterations
I am conducting some new trainings sessions to better understand certain influencing factors. Test of the model is done with my new detector. Now I go up to 10000 iterations to see, what happens.
#AI #YOLO #darknet #COCO #model #training #objectdetection #boundingbox #annotation #Python #Bash
I can now detect cats using my own cat models. For the detection I wrote a small Python detector on base of OpenCV using DNN as neural network. Now I can improve my model. NMS is influencing the result.
#AI #YOLO #COCO #model #training #detection #objectdetection #darknet #boundingbox #Python #Bash
I have a false positive rate of round about 4.5 % in the cat recognition. It is for me not clear, why e.g. a rabbit is wrong recognized.
#AI #YOLO #COCO #model #objectdetection #darknet #boundingbox #Python #Bash
This result has now amazed me. I didn't think that 100% recognition could be achieved or that 100% was achievable.
#AI #YOLO #COCO #model #objectdetection #darknet #boundingbox #Python #Bash
Interim result if I continue the training, in this case with some new images.
#AI #YOLO #COCO #model #objectdetection #darknet #boundingbox #Python #Bash
Best result using darknet in object recognition. The cat is recognized and the squirrel is no longer recognized. I am now continuing to pursue this approach. Now I need more tests.
#AI #YOLO #COCO #model #objectdetection #darknet #boundingbox #Python #Bash
Interim result. The new approach is currently more reliable than the old one. Left is the old recognition. Right is the new recognition. For the old one the squirrel was all the time a cat.
#AI #YOLO #COCO #model #objectdetection #darknet #boundingbox #Python #Bash
Here some results of my model with respect to the object cat. I need some fine tuning when a cat is small in comparison to the image size. I have to prepare some training images for this task.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
Here is my last modified model with false detections. Please note the reliability. Mainly exclusively the cat should be recognized.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
I created some images for testing to evaluate existing YOLOv5 models. I can use YOLO v5n-v5x so far next to my own model. Here are some false results. Everything is a bear. Funny. This is the base for testing my model.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
The recognition of cats with my YOLO model remains extremely high.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
These cats are no longer recognized with a threshold of 30 percent. Painting is also excluded now.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
Here comes the result of my approach to prevent false positives. I can target false positives for reduction. Comparison previous to new model. still needs to be proven.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
I was able to reduce false detections by using so-called ‘backgrounds’ in YOLO. I am now trying to use this in a targeted manner.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
Another nice wrong result in object recognition. Badger as bear.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
I will now show you an example of how the original used model differs from my sample images. Panther is bear. This is one reason why I am studying the training procedure.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
I will now show you an example of how the original used model differs from my sample images. Cat is bird.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
Now I will demonstrate the weakness of this approach with one single class. Here, a dog is recognized as cat. To make the model robust, I have to include four-legged animals in the model for differentiation purposes.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox
My model is actually only supposed to recognize domestic cats. But apparently, a cat is a cat, no matter how big it is.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox #cat
My model is actually only supposed to recognize domestic cats. But apparently, a cat is a cat, no matter how big it is.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox #cat
My model is actually only supposed to recognize domestic cats. But apparently, a cat is a cat, no matter how big it is.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox #cat
Here, I had expected recognition by the Yolo model.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox #cat
The Yolo model is now functioning better than anticipated. Here I used one of my AI images. This result was surprising to me.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox #cat
The Yolo model is now functioning better than anticipated. Here I used one of my AI images.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox #cat
I underestimated one thing when creating the YOLO model. The large number of cat breeds must be incorporated into the training. Here I used one of my photos and a free available for the test.
#AI #YOLO #model #objectdetection #Bash #Python #script #annotation #boundingbox #cat
Polygon annotation is the precision edge in object detection. Pixel-perfect accuracy, fewer false positives, smarter AI across industries.
Full guide: differ.blog/p/accurate-o...
#polygonannotation #objectdetection #Aitrainingdata #Datalabeling #machinelearning #computervision #Boundingbox #Ai
