I've downloaded the EfficientDet D0 512x512 model from the object detection API model zoo, downloaded the PASCAL VOC dataset and preprocessed it with the create_pascal_tf_record.py file. Next I took one of the config files and adjusted it to fit the architecture and VOC dataset. When evaluating the resulting network with the pascal_voc_detection_metrics it gives me a near zero mAP for the first class (airplane), the other classes are performing fine. I'm assuming one of my settings in the config file is wrong (pasted down below), why does this happen and how do i fix this?
model {
ssd {
inplace_batchnorm_update: true
freeze_batchnorm: false
num_classes: 20
add_background_class: false
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 {
}
}
encode_background_as_zeros: true
anchor_generator {
multiscale_anchor_generator {
min_level: 3
max_level: 7
anchor_scale: 4.0
aspect_ratios: [1.0, 2.0, 0.5]
scales_per_octave: 3
}
}
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 512
max_dimension: 512
pad_to_max_dimension: true
}
}
box_predictor {
weight_shared_convolutional_box_predictor {
depth: 64
class_prediction_bias_init: -4.6
conv_hyperparams {
force_use_bias: true
activation: SWISH
regularizer {
l2_regularizer {
weight: 0.00004
}
}
initializer {
random_normal_initializer {
stddev: 0.01
mean: 0.0
}
}
batch_norm {
scale: true
decay: 0.99
epsilon: 0.001
}
}
num_layers_before_predictor: 3
kernel_size: 3
use_depthwise: true
}
}
feature_extractor {
type: 'ssd_efficientnet-b0_bifpn_keras'
bifpn {
min_level: 3
max_level: 7
num_iterations: 3
num_filters: 64
}
conv_hyperparams {
force_use_bias: true
activation: SWISH
regularizer {
l2_regularizer {
weight: 0.00004
}
}
initializer {
truncated_normal_initializer {
stddev: 0.03
mean: 0.0
}
}
batch_norm {
scale: true,
decay: 0.99,
epsilon: 0.001,
}
}
}
loss {
classification_loss {
weighted_sigmoid_focal {
alpha: 0.25
gamma: 1.5
}
}
localization_loss {
weighted_smooth_l1 {
}
}
classification_weight: 1.0
localization_weight: 1.0
}
normalize_loss_by_num_matches: true
normalize_loc_loss_by_codesize: true
post_processing {
batch_non_max_suppression {
score_threshold: 1e-8
iou_threshold: 0.5
max_detections_per_class: 100
max_total_detections: 100
}
score_converter: SIGMOID
}
}
}
train_config: {
fine_tune_checkpoint: "oracle/efficientdet_d0/checkpoint/ckpt-0"
fine_tune_checkpoint_version: V2
fine_tune_checkpoint_type: "detection"
batch_size: 3
startup_delay_steps: 0
use_bfloat16: false
num_steps: 30000
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
random_scale_crop_and_pad_to_square {
output_size: 512
scale_min: 0.1
scale_max: 2.0
}
}
optimizer {
momentum_optimizer: {
learning_rate: {
cosine_decay_learning_rate {
learning_rate_base: 8e-2
total_steps: 30000
warmup_learning_rate: .001
warmup_steps: 2500
}
}
momentum_optimizer_value: 0.9
}
use_moving_average: false
}
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
update_trainable_variables: ["WeightSharedConvolutionalBoxPredictor"]
}
train_input_reader: {
label_map_path: "pascalVOC/pascal_label_map.pbtxt"
tf_record_input_reader {
input_path: "pascalVOC/pascal_train.record"
}
}
eval_config: {
metrics_set: "pascal_voc_detection_metrics"
use_moving_averages: false
batch_size: 1;
}
eval_input_reader: {
label_map_path: "pascalVOC/pascal_label_map.pbtxt"
shuffle: false
num_epochs: 1
tf_record_input_reader {
input_path: "pascalVOC/pascal_val.record"
}
}
There is a bug in the way pascal_voc_detection_metrics calculates the metric, fix can be found here