Im using TensorFlow Object Detection API with TensorFlow 2 and I have a problem with the training images that are displayed in TensorBoard like this:
even though the evaluation images look normally, so the problem is not in the data serialization.
To make sure that it is not done by image augmentation, I let only horizontal flip in the config file. See the full config file:
model {
center_net {
num_classes: 3
feature_extractor {
type: "resnet_v1_50_fpn"
}
image_resizer {
keep_aspect_ratio_resizer {
min_dimension: 512
max_dimension: 512
pad_to_max_dimension: true
}
}
object_detection_task {
task_loss_weight: 1.0
offset_loss_weight: 1.0
scale_loss_weight: 0.1
localization_loss {
l1_localization_loss {
}
}
}
object_center_params {
object_center_loss_weight: 1.0
min_box_overlap_iou: 0.7
max_box_predictions: 100
classification_loss {
penalty_reduced_logistic_focal_loss {
alpha: 2.0
beta: 4.0
}
}
}
}
}
train_config: {
batch_size: 5
num_steps: 10000
data_augmentation_options {
random_horizontal_flip {
}
}
optimizer {
adam_optimizer: {
epsilon: 1e-7 # Match tf.keras.optimizers.Adam's default.
learning_rate: {
cosine_decay_learning_rate {
learning_rate_base: 0.3e-3
total_steps: 10000
warmup_learning_rate: 0.3e-3
warmup_steps: 1000
}
}
}
use_moving_average: false
}
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
fine_tune_checkpoint_version: V2
fine_tune_checkpoint: "C:/ObjectDetection/FaceMaskDetection/Zoo/centernet_resnet50_v1_fpn_512x512_coco17_tpu-8/checkpoint/ckpt-0"
fine_tune_checkpoint_type: "fine_tune"
}
train_input_reader: {
label_map_path: "C:/ObjectDetection/FaceMaskDetection/Dataset/TFRecord/label_map.pbtxt"
tf_record_input_reader {
input_path: "C:/ObjectDetection/FaceMaskDetection/Dataset/TFRecord/train.record"
}
}
eval_config: {
metrics_set: "coco_detection_metrics"
use_moving_averages: false
batch_size: 1;
}
eval_input_reader: {
label_map_path: "C:/ObjectDetection/FaceMaskDetection/Dataset/TFRecord/label_map.pbtxt"
shuffle: false
num_epochs: 1
tf_record_input_reader {
input_path: "C:/ObjectDetection/FaceMaskDetection/Dataset/TFRecord/eval.record"
}
}
Im using centernet_resnet50_v1_fpn_512x512_coco17_tpu-8. The weird thing is that both loss and mAP look reasonable and I think that I wont be able to get those numbers with such bad training images. Is it just some visualization error?
