ultralytics
ultralytics copied to clipboard
ultralytics
Runnning on Tflite interpreter fails with different values
Search before asking
- [X] I have searched the YOLOv8 issues and found no similar bug report.
YOLOv8 Component
Detection
Bug
Converted YOLOv8s-seg.pt to tflite.
It seemes successful.
Input and output details are like this.
[{'name': 'inputs_0', 'index': 0, 'shape': array([ 1, 640, 640, 3], dtype=int32), 'shape_signature': array([ 1, 640, 640, 3], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}] [{'name': 'Identity', 'index': 498, 'shape': array([ 1, 160, 160, 32], dtype=int32), 'shape_signature': array([ 1, 160, 160, 32], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'Identity_1', 'index': 631, 'shape': array([ 1, 116, 8400], dtype=int32), 'shape_signature': array([ 1, 116, 8400], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]
So I used tflite interpreter to run this model on following code.
import torch
import cv2
from ultralytics.yolo.data.augment import LetterBox
from ultralytics.yolo.utils import ops
import copy
import numpy as np
def Preprocess(img):
original = copy.deepcopy(img)
img = LetterBox(640, True)(image=img)
img = img[::-1] #BGR2RGB
img = np.ascontiguousarray(img)
img = torch.from_numpy(img)
img = img.float()
img /= 255
return img,original
def Postprocess(preds, img, original):
preds = ops.non_max_suppression(preds,
0.25,
0.7,
agnostic=False,
max_det=100)
# at this point, preds values are different from Pytorch's
for i, pred in enumerate(preds):
shape = orig_img.shape
pred[:, :4] = ops.scale_boxes(img.shape[2:], pred[:, :4], shape).round()
return preds
image = cv2.imread(imgpath)
image,originalimg = Preprocess(image)
interpreter = tf.lite.Interpreter(model_path=model)
interpreter.allocate_tensors()
interpreter.set_tensor(input_details[0]['index'], img)
interpreter.invoke()
output_data = interpreter.get_tensor(output_details[1]['index']) #recieve [ 1, 116, 8400]
output_data = torch.from_numpy(output_data.astype(np.float32))
preds = postprocess(output_data,img,originalimg)
then, preds values are like this.
[tensor([[296.00000, 0.00000, 296.00000, 0.00000, 1.48874, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.46740, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.46368, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.46071, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.46027, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.42773, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.41953, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.40604, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.40517, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.39853, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.39473, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.37746, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.37366, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.36995, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.36336, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.35947, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.35938, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.35844, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.33753, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.33574, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.32853, 95.00000],
[296.00000, 297.00000, 296.00000, 297.00000, 1.32728, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.32687, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.32153, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.32110, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.32016, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.31914, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.31679, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.31636, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.31204, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.30827, 95.00000],
[296.00000, 297.00000, 296.00000, 297.00000, 1.30709, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.30572, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.30567, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.30566, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.30418, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.29971, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.29922, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.29898, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.29892, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.29326, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.29032, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.28821, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.28746, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.28691, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.27904, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.27829, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.27676, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.27225, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.27093, 105.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.26700, 95.00000],
[296.00000, 0.00000, 296.00000, 96.00000, 1.26693, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.26450, 105.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.26336, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.26283, 105.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.26265, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.25472, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.25440, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.25344, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.25097, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.24687, 95.00000],
[296.00000, 0.00000, 296.00000, 292.00000, 1.24384, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.24273, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.24143, 95.00000],
[296.00000, 297.00000, 296.00000, 297.00000, 1.24057, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.23750, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.23689, 105.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.23676, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.23587, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.23425, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.23425, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.23404, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.23129, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.23104, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.23092, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.23083, 95.00000],
[296.00000, 0.00000, 296.00000, 64.00000, 1.23058, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.23005, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.22921, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.22664, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.22435, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.22415, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.22353, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.22146, 105.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.22025, 95.00000],
[296.00000, 297.00000, 296.00000, 297.00000, 1.21873, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.21826, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.21644, 95.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.21607, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.21417, 105.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.21402, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.21330, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.21250, 95.00000],
[296.00000, 297.00000, 296.00000, 297.00000, 1.21248, 104.00000],
[296.00000, 0.00000, 296.00000, 297.00000, 1.21233, 95.00000],
[296.00000, 297.00000, 296.00000, 297.00000, 1.21095, 95.00000],
[296.00000, 297.00000, 296.00000, 297.00000, 1.21048, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.20775, 95.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.20686, 105.00000],
[296.00000, 0.00000, 296.00000, 0.00000, 1.20648, 95.00000]])]
This value is completely different from the original one. I'm not sure if this is caused by my code or converted model. Any advise would be very appriciated.
Thank you.
Environment
YOLOV8 Colaboratory-Python
Minimal Reproducible Example
No response
Additional
No response
Are you willing to submit a PR?
- [ ] Yes I'd like to help by submitting a PR!
👋 Hello there! We wanted to give you a friendly reminder that this issue has not had any recent activity and may be closed soon, but don't worry - you can always reopen it if needed. If you still have any questions or concerns, please feel free to let us know how we can help.
For additional resources and information, please see the links below:
- Docs: https://docs.ultralytics.com
- HUB: https://hub.ultralytics.com
- Community: https://community.ultralytics.com
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 YOLO 🚀 and Vision AI ⭐
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "Published by a pub.dev verified publisher"){kind=small}
@saka3324 you're welcome! We appreciate your patience and understanding. If you have any other questions or need further assistance, feel free to ask. We're here to help!
