About the ultralytics hub with tflite file with custom dataset

[khchan246810]

Hi, I am the developer for Android Studio object detection. I know I am new to this function, and I might going to use ultralytics hub. Since, the txt file is not required if the model has metadata based on your code. I am not sure it can work by simply exporting the tflite model without needing txt file into the Android Studio. If it did detect the objects, the name might be not labelled or something.

[surendramaran]

If the metadata is not present in tflite then the ultralytics python library or its environment itself cannot label the classes.
So the default classes will be class0, class1, ...... class999

This repo will get the labels, for sure if metadata is presented.
Let me know if it doesn't.

[khchan246810]

Ok thanks. Your YOLOv10 code got supported several YOLO versions such as YOLOv8, YOLOv9, and other versions? Or like YOLOv10 should have YOLOv10 tflite file, not other versions?

[surendramaran]

Yeah, they should be yolov10 only for yolov10.

Yolov8 and yolov9 you can use yolov8 or yolov9 repo.

[khchan246810]

Ok, thank you again.

[khchan246810]

Hi sir, I want to ask for your opinion. What do you recommended for your custom dataset in terms of epoch, batch size, patience, and image size? Let's say if I have over 1000 images for 80 classes.

[moodpanda]

may I ask what did you use to export yolov8n.pt to yolov8.tflite? im using the ultralytics YOLO.export() but it does not seem to work. thank you so much for your response im really struggling working with yolov8