YOLOv8 INT8 TFLite Inference Example (#7317)
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>pull/7390/head^2
Wamiq Raza
11 months ago
committed by
GitHub
parent
a6a2c256d4
commit
e76754eab0
3 changed files with 364 additions and 0 deletions
@ -0,0 +1,65 @@ |
||||
# YOLOv8 - Int8-TFLite Runtime |
||||
|
||||
Welcome to the YOLOv8 Int8 TFLite Runtime for efficient and optimized object detection project. This README provides comprehensive instructions for installing and using our YOLOv8 implementation. |
||||
|
||||
## Installation |
||||
|
||||
Ensure a smooth setup by following these steps to install necessary dependencies. |
||||
|
||||
### Installing Required Dependencies |
||||
|
||||
Install all required dependencies with this simple command: |
||||
|
||||
```bash |
||||
pip install -r requirements.txt |
||||
``` |
||||
|
||||
### Installing `tflite-runtime` |
||||
|
||||
To load TFLite models, install the `tflite-runtime` package using: |
||||
|
||||
```bash |
||||
pip install tflite-runtime |
||||
``` |
||||
|
||||
### Installing `tensorflow-gpu` (For NVIDIA GPU Users) |
||||
|
||||
Leverage GPU acceleration with NVIDIA GPUs by installing `tensorflow-gpu`: |
||||
|
||||
```bash |
||||
pip install tensorflow-gpu |
||||
``` |
||||
|
||||
**Note:** Ensure you have compatible GPU drivers installed on your system. |
||||
|
||||
### Installing `tensorflow` (CPU Version) |
||||
|
||||
For CPU usage or non-NVIDIA GPUs, install TensorFlow with: |
||||
|
||||
```bash |
||||
pip install tensorflow |
||||
``` |
||||
|
||||
## Usage |
||||
|
||||
Follow these instructions to run YOLOv8 after successful installation. |
||||
|
||||
Convert the YOLOv8 model to Int8 TFLite format: |
||||
|
||||
```bash |
||||
yolo export model=yolov8n.pt imgsz=640 format=tflite int8 |
||||
``` |
||||
|
||||
Locate the Int8 TFLite model in `yolov8n_saved_model`. Choose `best_full_integer_quant` or verify quantization at [Netron](https://netron.app/). Then, execute the following in your terminal: |
||||
|
||||
```bash |
||||
python main.py --model yolov8n_full_integer_quant.tflite --img image.jpg --conf-thres 0.5 --iou-thres 0.5 |
||||
``` |
||||
|
||||
Replace `best_full_integer_quant.tflite` with your model file's path, `image.jpg` with your input image, and adjust the confidence (conf-thres) and IoU thresholds (iou-thres) as necessary. |
||||
|
||||
### Output |
||||
|
||||
The output is displayed as annotated images, showcasing the model's detection capabilities: |
||||
|
||||
 |
||||
@ -0,0 +1,298 @@ |
||||
import argparse |
||||
|
||||
import cv2 |
||||
import numpy as np |
||||
from tflite_runtime import interpreter as tflite |
||||
|
||||
from ultralytics.utils import ASSETS, yaml_load |
||||
from ultralytics.utils.checks import check_yaml |
||||
|
||||
# Declare as global variables, can be updated based trained model image size |
||||
img_width = 640 |
||||
img_height = 640 |
||||
|
||||
|
||||
class LetterBox: |
||||
|
||||
def __init__(self, |
||||
new_shape=(img_width, img_height), |
||||
auto=False, |
||||
scaleFill=False, |
||||
scaleup=True, |
||||
center=True, |
||||
stride=32): |
||||
self.new_shape = new_shape |
||||
self.auto = auto |
||||
self.scaleFill = scaleFill |
||||
self.scaleup = scaleup |
||||
self.stride = stride |
||||
self.center = center # Put the image in the middle or top-left |
||||
|
||||
def __call__(self, labels=None, image=None): |
||||
"""Return updated labels and image with added border.""" |
||||
|
||||
if labels is None: |
||||
labels = {} |
||||
img = labels.get('img') if image is None else image |
||||
shape = img.shape[:2] # current shape [height, width] |
||||
new_shape = labels.pop('rect_shape', self.new_shape) |
||||
if isinstance(new_shape, int): |
||||
new_shape = (new_shape, new_shape) |
||||
|
||||
# Scale ratio (new / old) |
||||
r = min(new_shape[0] / shape[0], new_shape[1] / shape[1]) |
||||
if not self.scaleup: # only scale down, do not scale up (for better val mAP) |
||||
r = min(r, 1.0) |
||||
|
||||
# Compute padding |
||||
ratio = r, r # width, height ratios |
||||
new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r)) |
||||
dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1] # wh padding |
||||
if self.auto: # minimum rectangle |
||||
dw, dh = np.mod(dw, self.stride), np.mod(dh, self.stride) # wh padding |
||||
elif self.scaleFill: # stretch |
||||
dw, dh = 0.0, 0.0 |
||||
new_unpad = (new_shape[1], new_shape[0]) |
||||
ratio = new_shape[1] / shape[1], new_shape[0] / shape[0] # width, height ratios |
||||
|
||||
if self.center: |
||||
dw /= 2 # divide padding into 2 sides |
||||
dh /= 2 |
||||
|
||||
if shape[::-1] != new_unpad: # resize |
||||
img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR) |
||||
top, bottom = int(round(dh - 0.1)) if self.center else 0, int(round(dh + 0.1)) |
||||
left, right = int(round(dw - 0.1)) if self.center else 0, int(round(dw + 0.1)) |
||||
img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, |
||||
value=(114, 114, 114)) # add border |
||||
if labels.get('ratio_pad'): |
||||
labels['ratio_pad'] = (labels['ratio_pad'], (left, top)) # for evaluation |
||||
|
||||
if len(labels): |
||||
labels = self._update_labels(labels, ratio, dw, dh) |
||||
labels['img'] = img |
||||
labels['resized_shape'] = new_shape |
||||
return labels |
||||
else: |
||||
return img |
||||
|
||||
def _update_labels(self, labels, ratio, padw, padh): |
||||
"""Update labels.""" |
||||
|
||||
labels['instances'].convert_bbox(format='xyxy') |
||||
labels['instances'].denormalize(*labels['img'].shape[:2][::-1]) |
||||
labels['instances'].scale(*ratio) |
||||
labels['instances'].add_padding(padw, padh) |
||||
return labels |
||||
|
||||
|
||||
class Yolov8TFLite: |
||||
|
||||
def __init__(self, tflite_model, input_image, confidence_thres, iou_thres): |
||||
""" |
||||
Initializes an instance of the Yolov8TFLite class. |
||||
|
||||
Args: |
||||
tflite_model: Path to the TFLite model. |
||||
input_image: Path to the input image. |
||||
confidence_thres: Confidence threshold for filtering detections. |
||||
iou_thres: IoU (Intersection over Union) threshold for non-maximum suppression. |
||||
""" |
||||
|
||||
self.tflite_model = tflite_model |
||||
self.input_image = input_image |
||||
self.confidence_thres = confidence_thres |
||||
self.iou_thres = iou_thres |
||||
|
||||
# Load the class names from the COCO dataset |
||||
self.classes = yaml_load(check_yaml('coco128.yaml'))['names'] |
||||
|
||||
# Generate a color palette for the classes |
||||
self.color_palette = np.random.uniform(0, 255, size=(len(self.classes), 3)) |
||||
|
||||
def draw_detections(self, img, box, score, class_id): |
||||
""" |
||||
Draws bounding boxes and labels on the input image based on the detected objects. |
||||
|
||||
Args: |
||||
img: The input image to draw detections on. |
||||
box: Detected bounding box. |
||||
score: Corresponding detection score. |
||||
class_id: Class ID for the detected object. |
||||
|
||||
Returns: |
||||
None |
||||
""" |
||||
|
||||
# Extract the coordinates of the bounding box |
||||
x1, y1, w, h = box |
||||
|
||||
# Retrieve the color for the class ID |
||||
color = self.color_palette[class_id] |
||||
|
||||
# Draw the bounding box on the image |
||||
cv2.rectangle(img, (int(x1), int(y1)), (int(x1 + w), int(y1 + h)), color, 2) |
||||
|
||||
# Create the label text with class name and score |
||||
label = f'{self.classes[class_id]}: {score:.2f}' |
||||
|
||||
# Calculate the dimensions of the label text |
||||
(label_width, label_height), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1) |
||||
|
||||
# Calculate the position of the label text |
||||
label_x = x1 |
||||
label_y = y1 - 10 if y1 - 10 > label_height else y1 + 10 |
||||
|
||||
# Draw a filled rectangle as the background for the label text |
||||
cv2.rectangle(img, (int(label_x), int(label_y - label_height)), |
||||
(int(label_x + label_width), int(label_y + label_height)), color, cv2.FILLED) |
||||
|
||||
# Draw the label text on the image |
||||
cv2.putText(img, label, (int(label_x), int(label_y)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 1, cv2.LINE_AA) |
||||
|
||||
def preprocess(self): |
||||
""" |
||||
Preprocesses the input image before performing inference. |
||||
|
||||
Returns: |
||||
image_data: Preprocessed image data ready for inference. |
||||
""" |
||||
|
||||
# Read the input image using OpenCV |
||||
self.img = cv2.imread(self.input_image) |
||||
|
||||
print('image befor', self.img) |
||||
# Get the height and width of the input image |
||||
self.img_height, self.img_width = self.img.shape[:2] |
||||
|
||||
letterbox = LetterBox(new_shape=[img_width, img_height], auto=False, stride=32) |
||||
image = letterbox(image=self.img) |
||||
image = [image] |
||||
image = np.stack(image) |
||||
image = image[..., ::-1].transpose((0, 3, 1, 2)) |
||||
img = np.ascontiguousarray(image) |
||||
# n, h, w, c |
||||
image = img.astype(np.float32) |
||||
image_data = image / 255 |
||||
# Return the preprocessed image data |
||||
return image_data |
||||
|
||||
def postprocess(self, input_image, output): |
||||
""" |
||||
Performs post-processing on the model's output to extract bounding boxes, scores, and class IDs. |
||||
|
||||
Args: |
||||
input_image (numpy.ndarray): The input image. |
||||
output (numpy.ndarray): The output of the model. |
||||
|
||||
Returns: |
||||
numpy.ndarray: The input image with detections drawn on it. |
||||
""" |
||||
|
||||
boxes = [] |
||||
scores = [] |
||||
class_ids = [] |
||||
for i, pred in enumerate(output): |
||||
pred = np.transpose(pred) |
||||
for box in pred: |
||||
x, y, w, h = box[:4] |
||||
x1 = x - w / 2 |
||||
y1 = y - h / 2 |
||||
boxes.append([x1, y1, w, h]) |
||||
idx = np.argmax(box[4:]) |
||||
scores.append(box[idx + 4]) |
||||
class_ids.append(idx) |
||||
|
||||
indices = cv2.dnn.NMSBoxes(boxes, scores, self.confidence_thres, self.iou_thres) |
||||
|
||||
for i in indices: |
||||
# Get the box, score, and class ID corresponding to the index |
||||
box = boxes[i] |
||||
gain = min(img_width / self.img_width, img_height / self.img_height) |
||||
pad = round((img_width - self.img_width * gain) / 2 - |
||||
0.1), round((img_height - self.img_height * gain) / 2 - 0.1) |
||||
box[0] = (box[0] - pad[0]) / gain |
||||
box[1] = (box[1] - pad[1]) / gain |
||||
box[2] = box[2] / gain |
||||
box[3] = box[3] / gain |
||||
score = scores[i] |
||||
class_id = class_ids[i] |
||||
if scores[i] > 0.25: |
||||
print(box, score, class_id) |
||||
# Draw the detection on the input image |
||||
self.draw_detections(input_image, box, score, class_id) |
||||
|
||||
return input_image |
||||
|
||||
def main(self): |
||||
""" |
||||
Performs inference using a TFLite model and returns the output image with drawn detections. |
||||
|
||||
Returns: |
||||
output_img: The output image with drawn detections. |
||||
""" |
||||
|
||||
# Create an interpreter for the TFLite model |
||||
interpreter = tflite.Interpreter(model_path=self.tflite_model) |
||||
self.model = interpreter |
||||
interpreter.allocate_tensors() |
||||
|
||||
# Get the model inputs |
||||
input_details = interpreter.get_input_details() |
||||
output_details = interpreter.get_output_details() |
||||
|
||||
# Store the shape of the input for later use |
||||
input_shape = input_details[0]['shape'] |
||||
self.input_width = input_shape[1] |
||||
self.input_height = input_shape[2] |
||||
|
||||
# Preprocess the image data |
||||
img_data = self.preprocess() |
||||
img_data = img_data |
||||
# img_data = img_data.cpu().numpy() |
||||
# Set the input tensor to the interpreter |
||||
print(input_details[0]['index']) |
||||
print(img_data.shape) |
||||
img_data = img_data.transpose((0, 2, 3, 1)) |
||||
|
||||
scale, zero_point = input_details[0]['quantization'] |
||||
interpreter.set_tensor(input_details[0]['index'], img_data) |
||||
|
||||
# Run inference |
||||
interpreter.invoke() |
||||
|
||||
# Get the output tensor from the interpreter |
||||
output = interpreter.get_tensor(output_details[0]['index']) |
||||
scale, zero_point = output_details[0]['quantization'] |
||||
output = (output.astype(np.float32) - zero_point) * scale |
||||
|
||||
output[:, [0, 2]] *= img_width |
||||
output[:, [1, 3]] *= img_height |
||||
print(output) |
||||
# Perform post-processing on the outputs to obtain output image. |
||||
return self.postprocess(self.img, output) |
||||
|
||||
|
||||
if __name__ == '__main__': |
||||
# Create an argument parser to handle command-line arguments |
||||
parser = argparse.ArgumentParser() |
||||
parser.add_argument('--model', |
||||
type=str, |
||||
default='yolov8n_full_integer_quant.tflite', |
||||
help='Input your TFLite model.') |
||||
parser.add_argument('--img', type=str, default=str(ASSETS / 'bus.jpg'), help='Path to input image.') |
||||
parser.add_argument('--conf-thres', type=float, default=0.5, help='Confidence threshold') |
||||
parser.add_argument('--iou-thres', type=float, default=0.5, help='NMS IoU threshold') |
||||
args = parser.parse_args() |
||||
|
||||
# Create an instance of the Yolov8TFLite class with the specified arguments |
||||
detection = Yolov8TFLite(args.model, args.img, args.conf_thres, args.iou_thres) |
||||
|
||||
# Perform object detection and obtain the output image |
||||
output_image = detection.main() |
||||
|
||||
# Display the output image in a window |
||||
cv2.imshow('Output', output_image) |
||||
|
||||
# Wait for a key press to exit |
||||
cv2.waitKey(0) |
||||
Loading…
Reference in new issue