Jaime García Villena
commited on
Commit
·
7cd0692
1
Parent(s):
9684c94
add a way to test this tflite
Browse files- coco_labels.json +82 -0
- test_images/cat.jpg +0 -0
- test_yolob8_tflite.py +205 -0
coco_labels.json
ADDED
|
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"0": "person",
|
| 3 |
+
"1": "bicycle",
|
| 4 |
+
"2": "car",
|
| 5 |
+
"3": "motorcycle",
|
| 6 |
+
"4": "airplane",
|
| 7 |
+
"5": "bus",
|
| 8 |
+
"6": "train",
|
| 9 |
+
"7": "truck",
|
| 10 |
+
"8": "boat",
|
| 11 |
+
"9": "traffic light",
|
| 12 |
+
"10": "fire hydrant",
|
| 13 |
+
"11": "stop sign",
|
| 14 |
+
"12": "parking meter",
|
| 15 |
+
"13": "bench",
|
| 16 |
+
"14": "bird",
|
| 17 |
+
"15": "cat",
|
| 18 |
+
"16": "dog",
|
| 19 |
+
"17": "horse",
|
| 20 |
+
"18": "sheep",
|
| 21 |
+
"19": "cow",
|
| 22 |
+
"20": "elephant",
|
| 23 |
+
"21": "bear",
|
| 24 |
+
"22": "zebra",
|
| 25 |
+
"23": "giraffe",
|
| 26 |
+
"24": "backpack",
|
| 27 |
+
"25": "umbrella",
|
| 28 |
+
"26": "handbag",
|
| 29 |
+
"27": "tie",
|
| 30 |
+
"28": "suitcase",
|
| 31 |
+
"29": "frisbee",
|
| 32 |
+
"30": "skis",
|
| 33 |
+
"31": "snowboard",
|
| 34 |
+
"32": "sports ball",
|
| 35 |
+
"33": "kite",
|
| 36 |
+
"34": "baseball bat",
|
| 37 |
+
"35": "baseball glove",
|
| 38 |
+
"36": "skateboard",
|
| 39 |
+
"37": "surfboard",
|
| 40 |
+
"38": "tennis racket",
|
| 41 |
+
"39": "bottle",
|
| 42 |
+
"40": "wine glass",
|
| 43 |
+
"41": "cup",
|
| 44 |
+
"42": "fork",
|
| 45 |
+
"43": "knife",
|
| 46 |
+
"44": "spoon",
|
| 47 |
+
"45": "bowl",
|
| 48 |
+
"46": "banana",
|
| 49 |
+
"47": "apple",
|
| 50 |
+
"48": "sandwich",
|
| 51 |
+
"49": "orange",
|
| 52 |
+
"50": "broccoli",
|
| 53 |
+
"51": "carrot",
|
| 54 |
+
"52": "hot dog",
|
| 55 |
+
"53": "pizza",
|
| 56 |
+
"54": "donut",
|
| 57 |
+
"55": "cake",
|
| 58 |
+
"56": "chair",
|
| 59 |
+
"57": "couch",
|
| 60 |
+
"58": "potted plant",
|
| 61 |
+
"59": "bed",
|
| 62 |
+
"60": "dining table",
|
| 63 |
+
"61": "toilet",
|
| 64 |
+
"62": "tv",
|
| 65 |
+
"63": "laptop",
|
| 66 |
+
"64": "mouse",
|
| 67 |
+
"65": "remote",
|
| 68 |
+
"66": "keyboard",
|
| 69 |
+
"67": "cell phone",
|
| 70 |
+
"68": "microwave",
|
| 71 |
+
"69": "oven",
|
| 72 |
+
"70": "toaster",
|
| 73 |
+
"71": "sink",
|
| 74 |
+
"72": "refrigerator",
|
| 75 |
+
"73": "book",
|
| 76 |
+
"74": "clock",
|
| 77 |
+
"75": "vase",
|
| 78 |
+
"76": "scissors",
|
| 79 |
+
"77": "teddy bear",
|
| 80 |
+
"78": "hair drier",
|
| 81 |
+
"79": "toothbrush"
|
| 82 |
+
}
|
test_images/cat.jpg
ADDED
|
test_yolob8_tflite.py
ADDED
|
@@ -0,0 +1,205 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
# -*- coding: utf-8 -*-
|
| 3 |
+
"""
|
| 4 |
+
Created on Wed Oct 4 16:44:12 2023
|
| 5 |
+
|
| 6 |
+
@author: lin
|
| 7 |
+
"""
|
| 8 |
+
import glob
|
| 9 |
+
import sys
|
| 10 |
+
sys.path.append('../../..')
|
| 11 |
+
import os
|
| 12 |
+
import cv2
|
| 13 |
+
import json
|
| 14 |
+
import tensorflow as tf
|
| 15 |
+
import numpy as np
|
| 16 |
+
import matplotlib.pyplot as plt
|
| 17 |
+
# from utils.bbox_op import non_max_supression
|
| 18 |
+
|
| 19 |
+
def one_multiple_iou(box, boxes, box_area, boxes_area):
|
| 20 |
+
"""
|
| 21 |
+
Compute the intersection over union. 1 to multiple
|
| 22 |
+
Inputs:
|
| 23 |
+
box: numpy array with 1 box, ymin, xmin, ymax, xmax
|
| 24 |
+
boxes: numpy array with shape [N, 4] holding N boxes
|
| 25 |
+
|
| 26 |
+
Outputs:
|
| 27 |
+
a numpy array with shape [N*1] representing box areas
|
| 28 |
+
"""
|
| 29 |
+
|
| 30 |
+
# this is the iou of the box against all other boxes
|
| 31 |
+
assert boxes.shape[0] == boxes_area.shape[0]
|
| 32 |
+
|
| 33 |
+
ymin = np.maximum(box[0], boxes[:, 0]) # bottom
|
| 34 |
+
xmin = np.maximum(box[1], boxes[:, 1]) # left
|
| 35 |
+
ymax = np.minimum(box[2], boxes[:, 2]) # top
|
| 36 |
+
xmax = np.minimum(box[3], boxes[:, 3]) # rifht
|
| 37 |
+
|
| 38 |
+
# we ignore areas where the intersection side would be negative
|
| 39 |
+
# this is done by using maxing the side length by 0
|
| 40 |
+
intersections = np.maximum(ymax - ymin, 0) * np.maximum(xmax - xmin, 0)
|
| 41 |
+
# each union is then the box area
|
| 42 |
+
# added to each other box area minusing their intersection calculated above
|
| 43 |
+
unions = box_area + boxes_area - intersections
|
| 44 |
+
# element wise division
|
| 45 |
+
# if the intersection is 0, then their ratio is 0
|
| 46 |
+
ious = intersections / unions
|
| 47 |
+
return ious
|
| 48 |
+
def select_non_overlapping_bboxes(boxes, scores, iou_th):
|
| 49 |
+
ymin = boxes[:, 0]
|
| 50 |
+
ymax = boxes[:, 2]
|
| 51 |
+
xmin = boxes[:, 1]
|
| 52 |
+
xmax = boxes[:, 3]
|
| 53 |
+
|
| 54 |
+
# box coordinate ranges are inclusive-inclusive
|
| 55 |
+
areas = (ymax - ymin) * (xmax - xmin)
|
| 56 |
+
scores_indexes = list(np.argsort(scores))
|
| 57 |
+
keep_idx = []
|
| 58 |
+
while len(scores_indexes) > 0:
|
| 59 |
+
index = scores_indexes.pop()
|
| 60 |
+
keep_idx.append(index)
|
| 61 |
+
|
| 62 |
+
ious = one_multiple_iou(
|
| 63 |
+
boxes[index], boxes[scores_indexes], areas[index], areas[scores_indexes]
|
| 64 |
+
)
|
| 65 |
+
filtered_indexes = set((ious > iou_th).nonzero()[0])
|
| 66 |
+
|
| 67 |
+
scores_indexes = [
|
| 68 |
+
v for (i, v) in enumerate(scores_indexes) if i not in filtered_indexes
|
| 69 |
+
]
|
| 70 |
+
return keep_idx
|
| 71 |
+
def non_max_supression(boxes, scores, classes, iou_th):
|
| 72 |
+
"""
|
| 73 |
+
remover overlaped boundingboxes. Starting by the box with the highest score
|
| 74 |
+
if the iou is greater than the threshold, remove it, else keep it.
|
| 75 |
+
Inputs:
|
| 76 |
+
boxes: numpy array with shape [N, 4] holding N boxes。 [ymin, xmin, ymax, xmax]
|
| 77 |
+
scores: numpy array with shape [N, 1] holding the prediction score of each box
|
| 78 |
+
classes: numpy array with shape [N, 1] holding the class that each box belongs
|
| 79 |
+
iou_th: intersection over union threshold to consider the overlapping boxes have detect 2 objects
|
| 80 |
+
Output:
|
| 81 |
+
boxes, scores, classes with intersection over union ratio less than the threshold.
|
| 82 |
+
|
| 83 |
+
"""
|
| 84 |
+
# assert boxes.shape[0] == scores.shape[0]
|
| 85 |
+
if len(scores) == 0:
|
| 86 |
+
return boxes, scores, classes
|
| 87 |
+
keep_idx = select_non_overlapping_bboxes(boxes, scores, iou_th)
|
| 88 |
+
|
| 89 |
+
return boxes[keep_idx], scores[keep_idx], classes[keep_idx]
|
| 90 |
+
|
| 91 |
+
def preprocess(img_path):
|
| 92 |
+
image_np = cv2.imread(img_path)
|
| 93 |
+
|
| 94 |
+
image_np = center_crop(image_np)
|
| 95 |
+
|
| 96 |
+
image_np = cv2.resize(image_np, (640, 640))
|
| 97 |
+
#image_np = cv2.cvtColor(image_np, cv2.COLOR_BGR2RGB)
|
| 98 |
+
image_np = image_np.astype(float)
|
| 99 |
+
image_np /= 255.0
|
| 100 |
+
return image_np
|
| 101 |
+
|
| 102 |
+
def center_crop(img):
|
| 103 |
+
width, height = img.shape[1], img.shape[0]
|
| 104 |
+
|
| 105 |
+
crop_size = width if width < height else height
|
| 106 |
+
|
| 107 |
+
mid_x, mid_y = int(width/2), int(height/2)
|
| 108 |
+
|
| 109 |
+
cs2 = int(crop_size/2)
|
| 110 |
+
crop_img = img[mid_y-cs2:mid_y+cs2, mid_x-cs2:mid_x+cs2]
|
| 111 |
+
|
| 112 |
+
return crop_img
|
| 113 |
+
|
| 114 |
+
def postprocess_prediction(preds):
|
| 115 |
+
|
| 116 |
+
bboxes = preds[0][:4]
|
| 117 |
+
class_prob = preds[0, 4:]
|
| 118 |
+
classes = np.argmax(class_prob, axis=0)
|
| 119 |
+
scores = np.max(class_prob, axis=0)
|
| 120 |
+
# filter by threshold
|
| 121 |
+
valid_idx = np.where(scores>=min_th)[0]
|
| 122 |
+
bboxes = bboxes[:, valid_idx]
|
| 123 |
+
classes = classes[valid_idx]
|
| 124 |
+
scores = scores[valid_idx]
|
| 125 |
+
bboxes = bboxes.transpose()
|
| 126 |
+
bboxes = bboxes*640
|
| 127 |
+
xmin = bboxes[:,0]-bboxes[:,2]//2
|
| 128 |
+
xmax = bboxes[:,0]+bboxes[:,2]//2
|
| 129 |
+
ymin = bboxes[:,1]-bboxes[:,3]//2
|
| 130 |
+
ymax = bboxes[:,1]+bboxes[:,3]//2
|
| 131 |
+
xmin = np.clip(xmin, 0, 640)
|
| 132 |
+
ymin = np.clip(ymin, 0, 640)
|
| 133 |
+
|
| 134 |
+
bboxes = np.vstack([ymin, xmin, ymax, xmax])
|
| 135 |
+
bboxes = bboxes.transpose()
|
| 136 |
+
bboxes = bboxes.astype(int)
|
| 137 |
+
|
| 138 |
+
bboxes, scores, classes = non_max_supression(bboxes, scores, classes, iou_th=0.5)
|
| 139 |
+
idx = np.argsort(scores)[::-1]
|
| 140 |
+
bboxes = bboxes[idx]
|
| 141 |
+
classes = classes[idx]
|
| 142 |
+
scores = scores[idx]
|
| 143 |
+
return bboxes, classes, scores
|
| 144 |
+
|
| 145 |
+
def plot_prediction(image_np, bboxes, classes, scores, label_map):
|
| 146 |
+
color=(255,0,0)
|
| 147 |
+
thickness=5
|
| 148 |
+
font_scale=3
|
| 149 |
+
|
| 150 |
+
for i, box in enumerate(bboxes):
|
| 151 |
+
box = bboxes[i, :]
|
| 152 |
+
|
| 153 |
+
ymin, xmin, ymax, xmax = box
|
| 154 |
+
|
| 155 |
+
image_np = cv2.rectangle(image_np, (xmin, ymin), (xmax, ymax), color=color, thickness=thickness)
|
| 156 |
+
text_x = xmin - 10 if xmin > 20 else xmin + 10
|
| 157 |
+
text_y = ymin - 10 if ymin > 20 else ymin + 10
|
| 158 |
+
display_str = label_map[str(classes[i])]
|
| 159 |
+
|
| 160 |
+
cv2.putText(
|
| 161 |
+
image_np,
|
| 162 |
+
display_str,
|
| 163 |
+
(text_x, text_y),
|
| 164 |
+
cv2.FONT_HERSHEY_SIMPLEX,
|
| 165 |
+
font_scale,
|
| 166 |
+
color,
|
| 167 |
+
thickness,
|
| 168 |
+
)
|
| 169 |
+
plt.imshow(image_np)
|
| 170 |
+
plt.show()
|
| 171 |
+
|
| 172 |
+
def predict_yolo_tflite(intenpreter, image_np):
|
| 173 |
+
input_tensor = np.expand_dims(image_np, axis=0)
|
| 174 |
+
input_tensor = tf.convert_to_tensor(input_tensor, dtype=tf.float32)
|
| 175 |
+
|
| 176 |
+
interpreter.set_tensor(input_details[0]['index'], input_tensor.numpy())
|
| 177 |
+
|
| 178 |
+
interpreter.invoke()
|
| 179 |
+
preds = interpreter.get_tensor(output_details[0]['index'])
|
| 180 |
+
return preds
|
| 181 |
+
|
| 182 |
+
if __name__ == "__main__":
|
| 183 |
+
min_th = 0.1
|
| 184 |
+
labels_json = "coco_labels.json"
|
| 185 |
+
with open(labels_json) as f:
|
| 186 |
+
label_map = json.load(f)
|
| 187 |
+
img_path = "test_images"
|
| 188 |
+
saved_tflite = "tflite_model.tflite"
|
| 189 |
+
# load model
|
| 190 |
+
interpreter = tf.lite.Interpreter(model_path=saved_tflite)
|
| 191 |
+
interpreter.allocate_tensors()
|
| 192 |
+
input_details = interpreter.get_input_details()
|
| 193 |
+
output_details = interpreter.get_output_details()
|
| 194 |
+
print(input_details)
|
| 195 |
+
print(output_details)
|
| 196 |
+
images = glob.glob(os.path.join(img_path, "*"))
|
| 197 |
+
for img in images:
|
| 198 |
+
image_np = preprocess(img)
|
| 199 |
+
print(image_np.shape)
|
| 200 |
+
|
| 201 |
+
# image_np = np.array(Image.open(image_path))
|
| 202 |
+
preds = predict_yolo_tflite(interpreter, image_np)
|
| 203 |
+
bboxes, classes, scores = postprocess_prediction(preds)
|
| 204 |
+
|
| 205 |
+
plot_prediction(image_np, bboxes, classes, scores, label_map)
|