In [1]:
import os
import numpy as np
import tensorflow as tf
assert tf.__version__.startswith('2')
from tflite_model_maker import model_spec
from tflite_model_maker import image_classifier
from tflite_model_maker.config import ExportFormat
from tflite_model_maker.config import QuantizationConfig
from tflite_model_maker.image_classifier import DataLoader
import matplotlib.pyplot as plt
import tflite_model_maker
from PIL import Image, ImageOps
2021-11-07 16:35:43.845313: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory 2021-11-07 16:35:43.845345: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
In [2]:
image_path = 'bhavika_data/'
In [3]:
data = DataLoader.from_folder(image_path)
INFO:tensorflow:Load image with size: 327, num_label: 2, labels: fall, no-fall.
2021-11-07 16:35:46.782860: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory 2021-11-07 16:35:46.782884: W tensorflow/stream_executor/cuda/cuda_driver.cc:269] failed call to cuInit: UNKNOWN ERROR (303) 2021-11-07 16:35:46.782903: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:156] kernel driver does not appear to be running on this host (bhavika-ML): /proc/driver/nvidia/version does not exist 2021-11-07 16:35:46.783239: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
In [4]:
train_data, validation_data = data.split(0.8)
In [5]:
model = image_classifier.create(train_data, model_spec='efficientnet_lite0', validation_data=validation_data, epochs=5, batch_size=8)
print(model)
INFO:tensorflow:Retraining the models...
2021-11-07 16:36:09.169921: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:185] None of the MLIR Optimization Passes are enabled (registered 2)
Model: "sequential" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= hub_keras_layer_v1v2 (HubKer (None, 1280) 3413024 _________________________________________________________________ dropout (Dropout) (None, 1280) 0 _________________________________________________________________ dense (Dense) (None, 2) 2562 ================================================================= Total params: 3,415,586 Trainable params: 2,562 Non-trainable params: 3,413,024 _________________________________________________________________ None Epoch 1/5
/home/bhavika/.local/lib/python3.8/site-packages/keras/optimizer_v2/optimizer_v2.py:355: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead. warnings.warn(
32/32 [==============================] - 28s 824ms/step - loss: 0.7842 - accuracy: 0.6250 - val_loss: 0.7170 - val_accuracy: 0.7344 Epoch 2/5 32/32 [==============================] - 30s 969ms/step - loss: 0.4769 - accuracy: 0.8398 - val_loss: 0.9231 - val_accuracy: 0.7188 Epoch 3/5 32/32 [==============================] - 29s 920ms/step - loss: 0.4378 - accuracy: 0.8633 - val_loss: 0.6859 - val_accuracy: 0.8281 Epoch 4/5 32/32 [==============================] - 29s 940ms/step - loss: 0.3862 - accuracy: 0.9062 - val_loss: 0.5129 - val_accuracy: 0.7656 Epoch 5/5 32/32 [==============================] - 25s 811ms/step - loss: 0.3386 - accuracy: 0.9414 - val_loss: 0.5095 - val_accuracy: 0.8281 <tensorflow_examples.lite.model_maker.core.task.image_classifier.ImageClassifier object at 0x7ff2ff5e5eb0>
In [7]:
model.export(export_dir='ai_models', tflite_filename='john-custom-transfer-learned-fall-detect-model-efficientnet_lite.tflite')
INFO:tensorflow:Assets written to: /tmp/tmpcironj4j/assets
INFO:tensorflow:Assets written to: /tmp/tmpcironj4j/assets 2021-11-07 16:41:26.748529: I tensorflow/core/grappler/devices.cc:66] Number of eligible GPUs (core count >= 8, compute capability >= 0.0): 0 2021-11-07 16:41:26.748700: I tensorflow/core/grappler/clusters/single_machine.cc:357] Starting new session 2021-11-07 16:41:26.781690: I tensorflow/core/grappler/optimizers/meta_optimizer.cc:1137] Optimization results for grappler item: graph_to_optimize function_optimizer: Graph size after: 913 nodes (656), 923 edges (664), time = 17.393ms. function_optimizer: function_optimizer did nothing. time = 0.448ms. 2021-11-07 16:41:27.507539: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:351] Ignored output_format. 2021-11-07 16:41:27.507574: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:354] Ignored drop_control_dependency. fully_quantize: 0, inference_type: 6, input_inference_type: 3, output_inference_type: 3 WARNING:absl:For model inputs containing unsupported operations which cannot be quantized, the `inference_input_type` attribute will default to the original type.
INFO:tensorflow:Label file is inside the TFLite model with metadata.
INFO:tensorflow:Label file is inside the TFLite model with metadata.
INFO:tensorflow:Saving labels in /tmp/tmpdd__n6_t/labels.txt
INFO:tensorflow:Saving labels in /tmp/tmpdd__n6_t/labels.txt
INFO:tensorflow:TensorFlow Lite model exported successfully: ai_models/john-custom-transfer-learned-fall-detect-model-efficientnet_lite.tflite
INFO:tensorflow:TensorFlow Lite model exported successfully: ai_models/john-custom-transfer-learned-fall-detect-model-efficientnet_lite.tflite
Model Inference¶
In [8]:
def thumbnail(image=None, desired_size=None):
"""Resizes original image as close as possible to desired size.
Preserves aspect ratio of original image.
Does not modify the original image.
:Parameters:
----------
image : PIL.Image
Input Image for AI model detection.
desired_size : (width, height)
Size expected by the AI model.
:Returns:
-------
PIL.Image
Resized image fitting for the AI model input tensor.
"""
assert image
assert desired_size
thumb = image.copy()
w, h = desired_size
try:
# convert from numpy to native Python int type
# that PIL expects
if isinstance(w, np.generic):
w = w.item()
w = int(w)
h = h.item()
h = int(h)
thumb.thumbnail((w, h))
except Exception as e:
msg = (f"Exception in "
f"PIL.image.thumbnail(desired_size={desired_size}):"
f"type(width)={type(w)}, type(height)={type(h)}"
f"\n{e}"
)
raise RuntimeError(msg)
return thumb
def resize(image=None, desired_size=None):
"""Pad original image to exact size expected by input tensor.
Preserve aspect ratio to avoid confusing the AI model with
unnatural distortions. Pad the resulting image
with solid black color pixels to fill the desired size.
Do not modify the original image.
:Parameters:
----------
image : PIL.Image
Input Image sized to fit an input tensor but without padding.
Its possible that one size fits one tensor dimension exactly
but the other size is smaller than
the input tensor other dimension.
desired_size : (width, height)
Exact size expected by the AI model.
:Returns:
-------
PIL.Image
Resized image fitting exactly the AI model input tensor.
"""
assert image
assert desired_size
thumb = image.copy()
delta_w = desired_size[0] - thumb.size[0]
delta_h = desired_size[1] - thumb.size[1]
padding = (0, 0, delta_w, delta_h)
new_im = ImageOps.expand(thumb, padding)
assert new_im.size == desired_size
return new_im
In [9]:
interpreter = tf.lite.Interpreter(model_path="ai_models/john-custom-transfer-learned-fall-detect-model-efficientnet_lite.tflite")
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
input_shape = input_details[0]['shape']
width, height = 224, 224
input_shape
Out[9]:
array([ 1, 224, 224, 3], dtype=int32)
In [10]:
def prediction(path_image):
template_image_src = Image.open(path_image)
plt.imshow(template_image_src)
_thumbnail = thumbnail(image=template_image_src, desired_size=(width, height))
img = resize(image=_thumbnail, desired_size=(width, height))
img = np.expand_dims(img,0)
interpreter.set_tensor(input_details[0]['index'], img)
interpreter.invoke()
output_data = interpreter.get_tensor(output_details[0]['index'])
pred = "fall" if np.argmax(output_data[0])==0 else "no-fall"
return pred
In [11]:
path_image = "Images/fall_img_8.png"
pred = prediction(path_image)
pred
Out[11]:
'no-fall'
In [ ]: