How can I convert .pt weight to tflite?

[jaskiratsingh2000]

Hi @glenn-jocher I want to convert PyTorch weights to tflite how can I do that?

[glenn-jocher]

@jaskiratsingh2000 see tf.py:

yolov5/models/tf.py

Lines 1 to 12 in bbfafea

	# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
	"""
	TensorFlow/Keras and TFLite versions of YOLOv5
	Authored by https://github.com/zldrobit in PR https://github.com/ultralytics/yolov5/pull/1127
	Usage:
	$ python models/tf.py --weights yolov5s.pt --cfg yolov5s.yaml
	Export int8 TFLite models:
	$ python models/tf.py --weights yolov5s.pt --cfg models/yolov5s.yaml --tfl-int8 \
	--source path/to/images/ --ncalib 100

[jaskiratsingh2000]

@glenn-jocher Hey, Thanks for referring me to this issue. I appreciate that.

Can you also let me know that instead of "yolov5s.pt" weights can I add custom trained weights file as well like "best.pt" or "last.pt"

@glenn-jocher another quick question that will this get exported or what? Where can I find after that?

[glenn-jocher]

@jaskiratsingh2000 you can export any YOLOv5 model, that's the main purpose of the function. It wouldn't be much use if it only exported official models.

Exported models are placed in same parent directory as source model.

[jaskiratsingh2000]

But I couldn't find a way to export

…

On Sun, 29 Aug 2021, 5:40 pm Glenn Jocher, ***@***.***> wrote: @jaskiratsingh2000 <https://github.com/jaskiratsingh2000> you can export any YOLOv5 model, that's the main purpose of the function. It wouldn't be much use if it only exported official models. Exported models are placed in same parent directory as source model. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#4586 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACOBHK6O5ZRGGXK47ZM3X63T7IPR7ANCNFSM5DADDT4Q> .

[JNaranjo-Alcazar]

Following the usage indications:

python models/tf.py --weights runs/train/exp6/weights/best.pt --cfg yolov5s.yaml

I got:

Starting TensorFlow GraphDef export with TensorFlow 2.6.0...
TensorFlow GraphDef export failure: name 'keras_model' is not defined
Traceback (most recent call last):
  File "models/tf.py", line 491, in <module>
    tf.TensorSpec(keras_model.inputs[0].shape, keras_model.inputs[0].dtype))
NameError: name 'keras_model' is not defined

Starting TFLite export with TensorFlow 2.6.0...

TFLite export failure: name 'keras_model' is not defined
Traceback (most recent call last):
  File "models/tf.py", line 521, in <module>
    converter = tf.lite.TFLiteConverter.from_keras_model(keras_model)
NameError: name 'keras_model' is not defined

However, when running the same script as:

python models/tf.py --weights runs/train/exp6/weights/best.pt --cfg models/yolov5s.yaml 

the conversion is run correctly. This may be a typo in the documentation or an code issue...

Hope this feedback helps!

Thanks for the amazing work!

[glenn-jocher]

@JNaranjo-Alcazar thanks for the bug report! I'm not able to reproduce this, when I run the usage example in Colab everything works well:

# Setup
!git clone https://github.com/ultralytics/yolov5  # clone repo
%cd yolov5
%pip install -qr requirements.txt  # install dependencies

import torch
from IPython.display import Image, clear_output  # to display images

clear_output()
print(f"Setup complete. Using torch {torch.__version__} ({torch.cuda.get_device_properties(0).name if torch.cuda.is_available() else 'CPU'})")

# Reproduce
!python models/tf.py --weights yolov5s.pt --cfg yolov5s.yaml

[JNaranjo-Alcazar]

I had this bug using a Docker container (not the one available in the repository) with python 3.7 and installing all the requirements and tensorflow.

I think I should use Colab then.

Thanks for the quick reply

[Ronald-Kray]

@glenn-jocher
After running tf.py, I can't find the result file.
Where can I find tflite file?

[glenn-jocher]

@Ronald-Kray TF exports are handled by export.py now:

python export.py --weights yolov5s.pt --include tflite

[github-actions]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!

[kashishgoyal31]

PYTHONPATH=. python3 models/tf.py --weights weights/yolov5s.pt --cfg models/yolov5s.yaml --img 320 --tfl-int8 --source /data/dataset/coco/coco2017/train2017 --ncalib 100

I am trying to convert my trained model to tflite. Please guide about weights. Will I be using my model trained best.pt ?? and what is to be added in source?