Here, we will create SSD-MobileNet-V2 model for smart phone deteaction. We are going to use tensorflow-gpu 2.2 for this. I am using python version 3.7.7.
Create a workspace, for this create a directory tensorflow_model.
$mkdir tensorflow_model
$cd tensorflow_model
Create virtual environment for workspace.
$python3 -m venv env
$source env/bin/activate
Upgrade your pip. pip version must be greate than version 19.0
pip install --upgrade pip
Installling tensorflow
pip install tensorflow-gpu==2.2
Installing few other related libraries and dependencies
pip install pillow
pip install lxml
pip install jupyter
pip install matplotlib
Create directory images and images/train , images/test
$mkdir tensorflow_model/images
$mkdir tensorflow_model/images/train
$mkdir tensorflow_model/images/test
Upload training images and test images to images/training and images/test respectively
Dataset ( images ) labelling is required for the training purpose. Install labelImg for the same.
Clone labelImg repository
$git clone https://github.com/tzutalin/labelImg.git
$cd labelImg
Install the dependencies required for labelImg
$apt-get install pyqt5-dev-tools
$pip install -r requirements/requirements-linux-python3.txt
run the labelImg
python labelImg.py
For detail usages of lableImg please refer to its documentation.
Make sure to store the xml file and images in same directory for traning and test data. I have stored the xml information of each images of images/train and images/test in their directory respectively.
create directory data in workspace (tensorflow_model/data) and from workspace home folder run the xml-to-csv.py script
$mkdir data
$python xml-to-csv.python
This will create test_labels.csv and train_lables.csv in tensorflow/data directory. Please verify the images path mention in the csv files are correct and absolute path is mentioned.
clone tensorflow models repository in the workspace tensorflow_model and install related dependencies
$ git clone https://github.com/tensorflow/models.git
$ cd models/reasearch
$ protoc object_detection/protos/*.proto --python_out=.
$ pip install tf_slim
$ pip install pandas
Copy models/research/official and models/research/object_detection from tensorflow_model/models to workspace home directory that is tensorflow_model
cp -r models/research/official .
cp -r models/research/object_detection .
Generate the record file required for pipeline configuration
$python generate_tfrecord.py --csv_input=data/train_labels.csv --output_path=data/train.record
$python generate_tfrecord.py --csv_input=data/test_labels.csv --output_path=data/test.record
$mkdir training
$cd training
$touch object-detection.pbxt
add below line in the the file object-detection.pbxt
item {
id: 1
name: 'mobile'
}
Since I am creating the model for mobile object detection. I have given name mobile. you can add multiple items if you want to train your model with multiple objects like below
item {
id: 1
name: 'mobile'
}
item {
id: 2
name: 'laptop'
}
exit from training directory to workspace home directory tensorflow_model
Download the example model ssd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz and extract in workspace home directory
$wget http://download.tensorflow.org/models/object_detection/tf2/20200711/ssd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz
$tar -xvzf sd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz
change the pipeline config accordingly. The changes I made in the config file is higlighted in bold
model {
ssd {
num_classes: **1**
image_resizer {
fixed_shape_resizer {
height: 300
width: 300
}
}
feature_extractor {
type: "ssd_mobilenet_v2_keras"
depth_multiplier: 1.0
min_depth: 16
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.9999998989515007e-05
}
}
initializer {
truncated_normal_initializer {
mean: 0.0
stddev: 0.029999999329447746
}
}
activation: RELU_6
batch_norm {
decay: 0.9700000286102295
center: true
scale: true
epsilon: 0.0010000000474974513
train: true
}
}
override_base_feature_extractor_hyperparams: true
}
box_coder {
faster_rcnn_box_coder {
y_scale: 10.0
x_scale: 10.0
height_scale: 5.0
width_scale: 5.0
}
}
matcher {
argmax_matcher {
matched_threshold: 0.5
unmatched_threshold: 0.5
ignore_thresholds: false
negatives_lower_than_unmatched: true
force_match_for_each_row: true
use_matmul_gather: true
}
}
similarity_calculator {
iou_similarity {
}
}
box_predictor {
convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.9999998989515007e-05
}
}
initializer {
random_normal_initializer {
mean: 0.0
stddev: 0.009999999776482582
}
}
activation: RELU_6
batch_norm {
decay: 0.9700000286102295
center: true
scale: true
epsilon: 0.0010000000474974513
train: true
}
}
min_depth: 0
max_depth: 0
num_layers_before_predictor: 0
use_dropout: false
dropout_keep_probability: 0.800000011920929
kernel_size: 1
box_code_size: 4
apply_sigmoid_to_scores: false
class_prediction_bias_init: -4.599999904632568
}
}
anchor_generator {
ssd_anchor_generator {
num_layers: 6
min_scale: 0.20000000298023224
max_scale: 0.949999988079071
aspect_ratios: 1.0
aspect_ratios: 2.0
aspect_ratios: 0.5
aspect_ratios: 3.0
aspect_ratios: 0.33329999446868896
}
}
post_processing {
batch_non_max_suppression {
score_threshold: 9.99999993922529e-09
iou_threshold: 0.6000000238418579
max_detections_per_class: 100
max_total_detections: 100
use_static_shapes: false
}
score_converter: SIGMOID
}
normalize_loss_by_num_matches: true
loss {
localization_loss {
weighted_smooth_l1 {
delta: 1.0
}
}
classification_loss {
weighted_sigmoid_focal {
gamma: 2.0
alpha: 0.75
}
}
classification_weight: 1.0
localization_weight: 1.0
}
encode_background_as_zeros: true
normalize_loc_loss_by_codesize: true
inplace_batchnorm_update: true
freeze_batchnorm: false
}
}
train_config {
batch_size: **10**
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
ssd_random_crop {
}
}
sync_replicas: true
optimizer {
momentum_optimizer {
learning_rate {
cosine_decay_learning_rate {
learning_rate_base: 0.800000011920929
total_steps: 50000
warmup_learning_rate: 0.13333000242710114
warmup_steps: 2000
}
}
momentum_optimizer_value: 0.8999999761581421
}
use_moving_average: false
}
fine_tune_checkpoint: **"ssd_mobilenet_v2_320x320_coco17_tpu-8/checkpoint/ckpt-0"**
num_steps: 50000
startup_delay_steps: 0.0
replicas_to_aggregate: 8
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
fine_tune_checkpoint_type: **"detection"**
fine_tune_checkpoint_version: V2
}
train_input_reader {
label_map_path: **"training/object-detection.pbxt"**
tf_record_input_reader {
input_path: **"data/train.record"**
}
}
eval_config {
metrics_set: "coco_detection_metrics"
use_moving_averages: false
}
eval_input_reader {
label_map_path: **"training/object-detection.pbxt"**
shuffle: false
num_epochs: 1
tf_record_input_reader {
input_path: **"data/test.record"**
}
}
Copy and add model_main_tf2.py and exporter_main_v2.py file in the workspace home directory. I have modified the file to use GPU.
cp object_detection/model_main_tf2.py .
cp object_detection/exporter_main_v2.py .
Install dependendecies if any dependency is missing
pip install pycocotools
pip install scipy
pip install dataclasses
pip install pyyaml
Create directory trained-checkpoint. In this directory we will store all the checkpoint during the model training which will be later used to export model
$mkdir tensorflow_model/trained-checkpoint
Now train your model. This step will take sometime and it depends on how large your dataset. I have trained my model on around 100 images and it took around 1 hour.
$python model_main_tf2.py --pipeline_config_path=ssd_mobilenet_v2_320x320_coco17_tpu-8/pipeline.config --model_dir=trained-checkpoint --alsologtostderr --num_train_steps=50000 --sample_1_of_n_eval_examples=1 --num_eval_steps=1
create directory tensorflow_model/exported-model
$ mkdir tensorflow_model/exported-model
Run the below command to export your model in tensorflow_model/exported-model directory
$python exporter_main_v2.py --input_type image_tensor --pipeline_config_path ./ssd_mobilenet_v2_320x320_coco17_tpu-8/pipeline.config --trained_checkpoint_dir ./trained-checkpoint --output_directory exported-model/mobile-model
copy the object_detection_tutorial.ipynb in workspace home directory and run the jupter notebook
$ cp object_detection/colab_tutorials/object_detection_tutorial.ipynb .
$ jupyter notebook
Run the object_detection_tutorial.ipynb file. Make changes accordingly if you are using other directory structure.
