tensorflow - 为什么在我自己的数据集上使用 ssd_mobilenet_v1_pnp 的结果很差?
问题描述
张量流 1.12.0
我目前正在尝试使用我的数据集训练 SSD_Mobilenet_V1_pnp 模型(使用 COCO 预训练)。我的数据集有 490 张用于训练的图像和 210 张用于评估的图像,23 个类
label_map.pbtxt:
项目 { id:1 名称:'a' } 项目 { id:2 名称:'b' }
...
管道配置:
model {
ssd {
num_classes: 24
image_resizer {
fixed_shape_resizer {
height: 300
width: 300
}
}
feature_extractor {
type: "ssd_mobilenet_v1_ppn"
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.99999989895e-05
}
}
initializer {
random_normal_initializer {
mean: 0.0
stddev: 0.00999999977648
}
}
activation: RELU_6
batch_norm {
decay: 0.97000002861
center: true
scale: true
epsilon: 0.0010000000475
}
}
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 {
weight_shared_convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.99999989895e-05
}
}
initializer {
random_normal_initializer {
mean: 0.0
stddev: 0.00999999977648
}
}
activation: RELU_6
batch_norm {
decay: 0.97000002861
center: true
scale: true
epsilon: 0.0010000000475
train: true
}
}
depth: 512
num_layers_before_predictor: 1
kernel_size: 1
class_prediction_bias_init: -4.59999990463
share_prediction_tower: true
}
}
anchor_generator {
ssd_anchor_generator {
num_layers: 6
min_scale: 0.15000000596
max_scale: 0.949999988079
aspect_ratios: 1.0
aspect_ratios: 2.0
aspect_ratios: 0.5
aspect_ratios: 3.0
aspect_ratios: 0.333299994469
reduce_boxes_in_lowest_layer: false
}
}
post_processing {
batch_non_max_suppression {
score_threshold: 0.300000011921
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_focal {
gamma: 2.0
alpha: 0.75
}
}
classification_weight: 1.0
localization_weight: 1.5
}
encode_background_as_zeros: true
normalize_loc_loss_by_codesize: true
inplace_batchnorm_update: true
freeze_batchnorm: false
}
}
train_config {
batch_size: 512
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: 0.699999988079
total_steps: 50000
warmup_learning_rate: 0.13330000639
warmup_steps: 2000
}
}
momentum_optimizer_value: 0.899999976158
}
use_moving_average: false
}
fine_tune_checkpoint: "model.ckpt"
num_steps: 50000
startup_delay_steps: 0.0
replicas_to_aggregate: 8
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
from_detection_checkpoint: true
}
train_input_reader {
label_map_path: "annotations\label_map.pbtxt"
tf_record_input_reader {
input_path: "train.record"
}
}
eval_config {
num_examples: 210
max_evals: 10
metrics_set: "coco_detection_metrics"
use_moving_averages: false
}
eval_input_reader {
label_map_path: "annotations\label_map.pbtxt"
shuffle: false
num_epochs: 1
num_readers: 1
tf_record_input_reader {
input_path: "val.record"
}
}
火车:
python object_detection/model_main.py --logtostderr --pipeline_config_path=pipeline.config --model_dir=train
日志:
Accumulating evaluation results...
DONE (t=0.05s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.000
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
这是正常的吗?怎么解决?
解决方案
我注意到的几件事可能会对您有所帮助:
配置文件中的 num_classes 为 24,但您正在训练 23 个课程。
另请参阅您使用的是固定图像调整器,具体取决于照片的尺寸,这可能是一个问题,因为您没有保持纵横比。
由于您的数据集相当小,因此训练更少的步骤(20k)可能会略有改善。
如果这些都没有帮助,请考虑在配置文件中添加硬数据挖掘器参数以引入最少数量的负面示例。
推荐阅读
- python - 如何使一列的值与数据框中的另一列相匹配
- xamarin - 空白项目跨平台抛出错误“对象引用未设置为对象的实例”
- python - Python:如何使用嵌套的 min 和 max 函数在 Python 中进行递归?
- python - 具有不同默认图像的 django ManyToMany Image 字段
- postgresql - 如何在 psql 函数中包含隔离级别
- php - PHP:为什么乘以“1E3”返回double而不是int?
- java - 在Java中从字符串中删除空格
- android-layout - 从另一个视图更改 textview 文本
- excel - 如何修复具有两个扩展名 .xlsb.xlsm 的 Excel 文件
- excel - 是否有一个excel函数可以复制同一行,只增加一个单元格