Closed Faust404 closed 5 months ago
Curious on what kinda dataset and process was used to train the YOLO model. It works well with over 90% accuracy.
Creating an issue because I have no other way to reach out to the repo owners.
It was trained on a dataset of 1.9k images. I first labelled about 200-300 captchas myself and trained a model using those images. Then I used that model to label the rest of the images, making corrections myself wherever necessary. Then I trained a final model with all the 1.9k captchas labelled.
The model was trained using the YOLOv8x (68,162,238 params) model with a batch size of 16 for 224 epochs and with all the images resized to 320x320 pixels.
Curious on what kinda dataset and process was used to train the YOLO model. It works well with over 90% accuracy. Creating an issue because I have no other way to reach out to the repo owners.
It was trained on a dataset of 1.9k images. I first labelled about 200-300 captchas myself and trained a model using those images. Then I used that model to label the rest of the images, making corrections myself wherever necessary. Then I trained a final model with all the 1.9k captchas labelled.
The model was trained using the YOLOv8x (68,162,238 params) model with a batch size of 16 for 224 epochs and with all the images resized to 320x320 pixels.
Thanks for the reply. I was attempting something like this because I had tried template matching with openCV without much luck.
Were the images grayscaled or any other image processing done before training the model? Also, what was the accuracy you arrived at when you tested the model. Because I had arrived at 91 or so percent when i tested the model on about 100 captcha I had saved myself.
Curious on what kinda dataset and process was used to train the YOLO model. It works well with over 90% accuracy. Creating an issue because I have no other way to reach out to the repo owners.
It was trained on a dataset of 1.9k images. I first labelled about 200-300 captchas myself and trained a model using those images. Then I used that model to label the rest of the images, making corrections myself wherever necessary. Then I trained a final model with all the 1.9k captchas labelled. The model was trained using the YOLOv8x (68,162,238 params) model with a batch size of 16 for 224 epochs and with all the images resized to 320x320 pixels.
How is your model well over 500MB but the yolov8x is 131MB only? I trained one based on captcha too (3.4k images~) but it just a little over 134MB.
What kinda accuracy were you able to obtain with your model. Also any chance you have a link to the dataset and model you trained?
Curious on what kinda dataset and process was used to train the YOLO model. It works well with over 90% accuracy. Creating an issue because I have no other way to reach out to the repo owners.
It was trained on a dataset of 1.9k images. I first labelled about 200-300 captchas myself and trained a model using those images. Then I used that model to label the rest of the images, making corrections myself wherever necessary. Then I trained a final model with all the 1.9k captchas labelled. The model was trained using the YOLOv8x (68,162,238 params) model with a batch size of 16 for 224 epochs and with all the images resized to 320x320 pixels.
How is your model well over 500MB but the yolov8x is 131MB only? I trained one based on captcha too (3.4k images~) but it just a little over 134MB.
What kinda accuracy were you able to obtain with your model. Also any chance you have a link to the dataset and model you trained?
94-95% accuracy. Dataset link https://drive.google.com/drive/folders/1--rPnwhIIn6zMyR3Trl5Z6aUoenp0Px7
Curious on what kinda dataset and process was used to train the YOLO model. It works well with over 90% accuracy. Creating an issue because I have no other way to reach out to the repo owners.
It was trained on a dataset of 1.9k images. I first labelled about 200-300 captchas myself and trained a model using those images. Then I used that model to label the rest of the images, making corrections myself wherever necessary. Then I trained a final model with all the 1.9k captchas labelled. The model was trained using the YOLOv8x (68,162,238 params) model with a batch size of 16 for 224 epochs and with all the images resized to 320x320 pixels.
How is your model well over 500MB but the yolov8x is 131MB only? I trained one based on captcha too (3.4k images~) but it just a little over 134MB.
What kinda accuracy were you able to obtain with your model. Also any chance you have a link to the dataset and model you trained?
94-95% accuracy. Dataset link https://drive.google.com/drive/folders/1--rPnwhIIn6zMyR3Trl5Z6aUoenp0Px7
Thanks for the dataset. I did already get it from lufy who I contacted through discord and I trained my own version of the model as well. I trained a nano model because I wanted to see how small I can go without losing performance and it's performing surprisingly well. I was able to obtain over 95% accuracy with the yolov8n model and it's only 6MB in size. It was trained with 3.2k images.
Maybe you can test the model and let me know how it goes. I have uploaded my dataset and the nano model in that same gdrive link. Also please do check your email, I have contacted you there as well.
Curious on what kinda dataset and process was used to train the YOLO model. It works well with over 90% accuracy.
Creating an issue because I have no other way to reach out to the repo owners.