Skip to main content

Yolo App - Train a Model with Tensorflow

Shenzhen, China

  1. Prepare your Images and get Data
  2. Train your Tensorflow Model
  3. Use your Model to do Predictions
  4. Use Tesseract to Read Number Plates
  5. Flask Web Application
  6. Yolo v5 - Data Prep

Prepare the Neural Network​

When running the following code I kept getting this warning:

Memory Error in Tensorflow: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero

Running the following command from my console to set each device to 0 got rid of the warning:

for a in /sys/bus/pci/devices/*; do echo 0 | sudo tee -a $a/numa_node; done
import tensorflow as tf
from tensorflow.keras.applications import MobileNetV2, InceptionV3, InceptionResNetV2
from tensorflow.keras.layers import Dense, Dropout, Flatten
from tensorflow.keras.models import Model

# Inception ResNet for 224x224 3-channel image source
inception_resnet = InceptionResNetV2(weights="imagenet", include_top=False, input_tensor=Input(shape=(224, 224, 3)))
inception_resnet.trainable=False
headmodel = inception_resnet.output
headmodel = Flatten()(headmodel)
# Build neural network layers
headmodel = Dense(500,activation='relu')(headmodel)
headmodel = Dense(250,activation='relu')(headmodel)
headmodel = Dense(4, activation='sigmoid')(headmodel)
# Define inputs / outputs for your model
model = Model(inception_resnet.input, outputs=headmodel)
# Compile the model
model.compile(loss='mse', optimizer=tf.keras.optimizers.Adam(learning_rate=1e-4))
2022-03-05 16:55:55.576063: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 4607 MB memory:  -> device: 0, name: NVIDIA GeForce GTX 1060 6GB, pci bus id: 0000:01:00.0, compute capability: 6.1

Train your Model​

from tensorflow.keras.callbacks import TensorBoard
tfb = TensorBoard('object_detection')
history = model.fit(x=x_train, y=y_train, batch_size=10, epochs=100, validation_data=(x_test, y_test), callbacks=[tfb])

With tensorflow-gpu on an Nvidia GTX 1060 I am getting training cycles of 2-3s:

Epoch 1/100

2022-03-05 17:44:41.082245: I tensorflow/stream_executor/cuda/cuda_dnn.cc:368] Loaded cuDNN version 8301

18/18 [==============================] - 13s 289ms/step - loss: 0.1179 - val_loss: 0.1214
Epoch 2/100
18/18 [==============================] - 2s 128ms/step - loss: 0.1204 - val_loss: 0.1146
Epoch 3/100
18/18 [==============================] - 2s 129ms/step - loss: 0.1223 - val_loss: 0.1235
...
Epoch 499/500
20/20 [==============================] - 3s 136ms/step - loss: 3.2455e-05 - val_loss: 0.0042
Epoch 500/500
20/20 [==============================] - 3s 137ms/step - loss: 2.7786e-05 - val_loss: 0.0042

To save the model run:

model.save('../models/object_detection.h5')

I can use Tensorboard to check the results - run the following command from your console and point it to your training data folder:

tensorboard --logdir=object_detection

TensorBoard 2.8.0 at http://localhost:6006/ (Press CTRL+C to quit)

Yolo App Tensorflow Model