我在这个领域有点新手。所以,如果这是非常基本的事情,我提前道歉。我目前正在尝试按照此链接https://colab.research.google.com/github/google-coral/tutorials/blob/master/retrain_detection_qat_tf1.ipynb#scrollTo=6RxtslKJf2td 使用 TensorFlow 对象检测 API 训练量化感知模型. 我使用 mobilenet_v2_quant_aware 作为我的模型。图像是 1000 个,有 800 个火车和 200 个火车。我面临的问题是损失没有收敛。这是管道:
model {
SSD {
num_classes: 1
image_resizer {
fixed_shape_resizer {
height: 300
width: 300
}
}
feature_extractor {
type: "ssd_mobilenet_v2"
depth_multiplier: 1.0
min_depth: 16
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.99999989895e-05
}
}
initializer {
truncated_normal_initializer {
mean: 0.0
stddev: 0.0299999993294
}
}
activation: RELU_6
batch_norm {
decay: 0.999700009823
center: true
scale: true
epsilon: 0.0010000000475
train: 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
}
}
similarity_calculator {
iou_similarity {
}
}
box_predictor {
convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.99999989895e-05
}
}
initializer {
truncated_normal_initializer {
mean: 0.0
stddev: 0.0299999993294
}
}
activation: RELU_6
batch_norm {
decay: 0.999700009823
center: true
scale: true
epsilon: 0.0010000000475
train: true
}
}
min_depth: 0
max_depth: 0
num_layers_before_predictor: 0
use_dropout: false
dropout_keep_probability: 0.800000011921
kernel_size: 1
box_code_size: 4
apply_sigmoid_to_scores: false
}
}
anchor_generator {
ssd_anchor_generator {
num_layers: 6
min_scale: 0.20000000298
max_scale: 0.949999988079
aspect_ratios: 1.0
aspect_ratios: 2.0
aspect_ratios: 0.5
aspect_ratios: 3.0
aspect_ratios: 0.333299994469
}
}
post_processing {
batch_non_max_suppression {
score_threshold: 9.99999993923e-09
iou_threshold: 0.600000023842
max_detections_per_class: 100
max_total_detections: 100
}
score_converter: SIGMOID
}
normalize_loss_by_num_matches: true
loss {
localization_loss {
weighted_smooth_l1 {
}
}
classification_loss {
weighted_sigmoid {
}
}
hard_example_miner {
num_hard_examples: 3000
iou_threshold: 0.990000009537
loss_type: CLASSIFICATION
max_negatives_per_positive: 3
min_negatives_per_image: 3
}
classification_weight: 1.0
localization_weight: 1.0
}
}
}
train_config {
batch_size: 8
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
ssd_random_crop {
}
}
optimizer {
rms_prop_optimizer {
learning_rate {
exponential_decay_learning_rate {
initial_learning_rate: 0.00500000018999
decay_steps: 10000
decay_factor: 0.949999988079
}
}
momentum_optimizer_value: 0.899999976158
decay: 0.899999976158
epsilon: 1.0
}
}
fine_tune_checkpoint: "/home/syslab3/Documents/TF_9/tod_tf1/pre_trained/ssd_mobilenet_v2_quantized_300x300_coco_2019_01_03/model.ckpt"
from_detection_checkpoint: true
num_steps: 50000
}
train_input_reader {
label_map_path: "/home/syslab3/Documents/TF_9/tod_tf1/training/objectDetection/fire_label.pbtxt"
tf_record_input_reader {
input_path: "/home/syslab3/Documents/TF_9/tod_tf1/training/objectDetection/train.tfrecord"
}
}
eval_config {
num_examples: 8000
metrics_set: "coco_detection_metrics"
use_moving_averages: true
include_metrics_per_category: true
}
eval_input_reader {
label_map_path: "/home/syslab3/Documents/TF_9/tod_tf1/training/objectDetection/fire_label.pbtxt"
shuffle: false
num_readers: 1
tf_record_input_reader {
input_path: "/home/syslab3/Documents/TF_9/tod_tf1/training/objectDetection/test.tfrecord"
}
}
graph_rewriter {
quantization {
delay: 48000
weight_bits: 8
activation_bits: 8
}
}
您能否建议一些我可以做的更改,或者如果导致问题,我应该增加我的数据集吗?
我尝试了不同的批量大小、学习率、衰减步骤和衰减因子,但没有帮助!
TensorFlow-gpu 1.15 GPU = Nvidia gtx 780