I've been playing with the Tensorflow Object Detection API 2(TF OD 2) in these days, I'm using the git head commit ce3b7227. My aim is to find the most suitable model for my custom dataset, by using the existent DL Architecture present in theTensorFlow 2 Model Zoo. I've generated my TF Records with the following tutorial of Roboflow and I have been training it with my Laptop and Google Colab, in GPU Mode.
I've found this amazing Roboflow's Colab Notebook, while I've tried to reproduce the same steps with my dataset, by using the models/research/object_detection/model_main_tf2.py, unluckly for me, the training script always ends before it started to iterate. It didn't show any Python Error and also it show some warnings as usual. The complete output is in my Colab Notebook
I'm fine-tuning the model with the following commands.
PIPELINE_CONFIG_PATH=models/ssd_resnet152_v1_fpn_640x640_coco17_tpu-8/pipeline.config; MODEL_DIR=training/; NUM_TRAIN_STEPS=10000; SAMPLE_1_OF_N_EVAL_EXAMPLES=1;
python models/research/object_detection/model_main_tf2.py --model_dir=$MODEL_DIR --num_train_steps=$NUM_TRAIN_STEPS --sample_1_of_n_eval_examples=$SAMPLE_1_OF_N_EVAL_EXAMPLES --pipeline_config_path=$PIPELINE_CONFIG_PATH --alsologtostderr
This is my pipeline.config file
model {
ssd {
num_classes: 90
image_resizer {
fixed_shape_resizer {
height: 640
width: 640
}
}
feature_extractor {
type: "ssd_resnet152_v1_fpn_keras"
depth_multiplier: 1.0
min_depth: 16
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 0.00039999998989515007
}
}
initializer {
truncated_normal_initializer {
mean: 0.0
stddev: 0.029999999329447746
}
}
activation: RELU_6
batch_norm {
decay: 0.996999979019165
scale: true
epsilon: 0.0010000000474974513
}
}
override_base_feature_extractor_hyperparams: true
fpn {
min_level: 3
max_level: 7
}
}
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 {
weight_shared_convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 0.00039999998989515007
}
}
initializer {
random_normal_initializer {
mean: 0.0
stddev: 0.009999999776482582
}
}
activation: RELU_6
batch_norm {
decay: 0.996999979019165
scale: true
epsilon: 0.0010000000474974513
}
}
depth: 256
num_layers_before_predictor: 4
kernel_size: 3
class_prediction_bias_init: -4.599999904632568
}
}
anchor_generator {
multiscale_anchor_generator {
min_level: 3
max_level: 7
anchor_scale: 4.0
aspect_ratios: 1.0
aspect_ratios: 2.0
aspect_ratios: 0.5
scales_per_octave: 2
}
}
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 {
}
}
classification_loss {
weighted_sigmoid_focal {
gamma: 2.0
alpha: 0.25
}
}
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: 8
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
random_crop_image {
min_object_covered: 0.0
min_aspect_ratio: 0.75
max_aspect_ratio: 3.0
min_area: 0.75
max_area: 1.0
overlap_thresh: 0.0
}
}
sync_replicas: true
optimizer {
momentum_optimizer {
learning_rate {
cosine_decay_learning_rate {
learning_rate_base: 0.03999999910593033
total_steps: 25000
warmup_learning_rate: 0.013333000242710114
warmup_steps: 2000
}
}
momentum_optimizer_value: 0.8999999761581421
}
use_moving_average: false
}
fine_tune_checkpoint_version: V2
fine_tune_checkpoint: "models/ssd_resnet152_v1_fpn_640x640_coco17_tpu-8/checkpoint/ckpt-0"
num_steps: 25000
startup_delay_steps: 0.0
replicas_to_aggregate: 8
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
fine_tune_checkpoint_type: "classification"
use_bfloat16: true
}
train_input_reader {
label_map_path: "datasets/UrbanTracker/urban_tracker_label_map.pbtxt"
tf_record_input_reader {
input_path: "datasets/UrbanTracker/urban_tracker_train.record"
}
}
eval_config {
metrics_set: "coco_detection_metrics"
use_moving_averages: false
}
eval_input_reader {
label_map_path: "datasets/UrbanTracker/urban_tracker_label_map.pbtxt"
shuffle: false
num_epochs: 1
tf_record_input_reader {
input_path: "datasets/UrbanTracker/urban_tracker_test.record"
}
}
And this is how my model's directory looks like.
.
├── datasets
│ ├── raccoon
│ │ ├── raccoon_label_map.pbtxt
│ │ ├── raccoon_test.record
│ │ └── raccoon_train.record
│ ├── readme.md
│ └── UrbanTracker
│ ├── labels_urbantracker.txt
│ ├── urban_tracker_label_map.pbtxt
│ ├── urban_tracker_test.record
│ └── urban_tracker_train.record
├── __main__.py
├── models
│ ├── AUTHORS
│ ├── efficientdet_d1_coco17_tpu-32
│ │ ├── checkpoint
│ │ │ ├── checkpoint
│ │ │ ├── ckpt-0.data-00000-of-00001
│ │ │ └── ckpt-0.index
│ │ ├── pipeline.config
│ │ ├── saved_model
│ │ │ ├── assets
│ │ │ ├── saved_model.pb
│ │ │ └── variables
│ │ │ ├── variables.data-00000-of-00001
│ │ │ └── variables.index
│ ├── faster_rcnn_resnet101_v1_640x640_coco17_tpu-8
│ │ ├── checkpoint
│ │ │ ├── checkpoint
│ │ │ ├── ckpt-0.data-00000-of-00001
│ │ │ └── ckpt-0.index
│ │ ├── pipeline.config
│ │ ├── saved_model
│ │ │ ├── saved_model.pb
│ │ │ └── variables
│ │ │ ├── variables.data-00000-of-00001
│ │ │ └── variables.index
│ └── ssd_resnet152_v1_fpn_640x640_coco17_tpu-8
│ ├── checkpoint
│ │ ├── checkpoint
│ │ ├── ckpt-0.data-00000-of-00001
│ │ └── ckpt-0.index
│ ├── pipeline.config
│ ├── saved_model
│ │ ├── assets
│ │ ├── saved_model.pb
│ │ └── variables
│ │ ├── variables.data-00000-of-00001
│ │ └── variables.index
├── tools
│ ├── parse_polytrack.py
│ ├── polytrack_csv_to_tfrecord.py
│ ├── raccoon_labels_test.csv
│ ├── raccoon_labels_train.csv
│ ├── split_dataset.py
│ ├── urban_tracker_test.csv
│ └── urban_tracker_train.csv
I've converted my Dataset to TFRecord with the TF v1 and v2 API. Also, I've been playing with different training parameters without luck. To check my dataset, in case I've generated it wrongly, I tried with another dataset, the basic Raccoon Dataset but I got the same result.
Thank you for your attention.
Solved: For models such as efficientdet_d1_coco17_tpu-32 just change in the pipeline.config the parameter from fine_tune_checkpoint_type: "classification" to fine_tune_checkpoint_type: "detection", check TF Github