TensorFlow 对象检测 API 过拟合
Posted
技术标签:
【中文标题】TensorFlow 对象检测 API 过拟合【英文标题】:Tensorflow object detection API overfitting 【发布时间】:2021-10-18 07:44:52 【问题描述】:我正在借助以下教程使用 Tensoflow 对象检测 API 创建树检测器:https://www.youtube.com/watch?v=a1br6gW-8Ss
我的问题是模型过度拟合了训练集。我怎样才能避免它? 我的训练集由 130 张图像组成,图像中平均标记了 4 棵树。共有7种。 我使用的模型是:在 COCO 数据集上预训练的“Faster R-CNN ResNet50 V1 640x640”..(链接如下) https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md 在教程中,导师使用以下模型“EfficientDet D0 512x512”,但该模型也过拟合了我的数据集,所以我尝试使用“Faster R-CNN ResNet50 V1 640x640”。
我用来减少过拟合的技术是
-
辍学(率=0.675)
数据增强(random_crop)
学习率低(learning_rate_base: .008 , total_steps: 25000 .warmup_learning_rate: .0001)
原始图像尺寸为 860x860,我将它们调整为 640x640,并在 xml 中更改了边界框坐标。
我应该在模型配置中添加或更改什么以避免过度拟合?
这是我的模型配置文件:
# Faster R-CNN with Resnet-50 (v1) with 640x640 input resolution
# Trained on COCO, initialized from Imagenet classification checkpoint
#
# Train on TPU-8
#
# Achieves 29.3 mAP on COCO17 Val
model
faster_rcnn
num_classes: 7
image_resizer
keep_aspect_ratio_resizer
min_dimension: 640
max_dimension: 640
pad_to_max_dimension: true
feature_extractor
type: 'faster_rcnn_resnet50_keras'
batch_norm_trainable: true
first_stage_anchor_generator
grid_anchor_generator
scales: [0.25, 0.5, 1.0, 2.0]
aspect_ratios: [0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5]
height_stride: 8
width_stride: 8
first_stage_box_predictor_conv_hyperparams
op: CONV
regularizer
l2_regularizer
weight: 0.0
initializer
truncated_normal_initializer
stddev: 0.01
first_stage_nms_score_threshold: 0.0
first_stage_nms_iou_threshold: 0.7
first_stage_max_proposals: 300
first_stage_localization_loss_weight: 2.0
first_stage_objectness_loss_weight: 1.0
initial_crop_size: 14
maxpool_kernel_size: 2
maxpool_stride: 2
second_stage_box_predictor
mask_rcnn_box_predictor
use_dropout: true
dropout_keep_probability: 0.675
fc_hyperparams
op: FC
regularizer
l2_regularizer
weight: 0.0
initializer
variance_scaling_initializer
factor: 1.0
uniform: true
mode: FAN_AVG
share_box_across_classes: true
second_stage_post_processing
batch_non_max_suppression
score_threshold: 0.2
iou_threshold: 0.6
max_detections_per_class: 100
max_total_detections: 300
score_converter: SOFTMAX
second_stage_localization_loss_weight: 2.0
second_stage_classification_loss_weight: 1.0
use_static_shapes: true
use_matmul_crop_and_resize: true
clip_anchors_to_image: true
use_static_balanced_label_sampler: true
use_matmul_gather_in_matcher: true
train_config:
batch_size: 4
sync_replicas: true
startup_delay_steps: 0
replicas_to_aggregate: 8
num_steps: 25000
optimizer
momentum_optimizer:
learning_rate:
cosine_decay_learning_rate
learning_rate_base: .008
total_steps: 25000
warmup_learning_rate: .0001
warmup_steps: 2000
momentum_optimizer_value: 0.9
use_moving_average: false
fine_tune_checkpoint_version: V2
fine_tune_checkpoint: "faster_rcnn_resnet50_v1_640x640_coco17_tpu-8/checkpoint/ckpt-0"
fine_tune_checkpoint_type: "detection"
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
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
use_bfloat16: true # works only on TPUs
train_input_reader:
label_map_path: "images/labelmap.pbtxt"
tf_record_input_reader
input_path: "train.record"
eval_config:
metrics_set: "coco_detection_metrics"
use_moving_averages: false
batch_size: 1;
eval_input_reader:
label_map_path: "images/labelmap.pbtxt"
shuffle: false
num_epochs: 1
tf_record_input_reader
input_path: "test.record"
【问题讨论】:
【参考方案1】:除了您为减少过拟合而采取的措施外,您还可以添加一些措施并对现有措施进行一些更改。
确保很好地对数据进行混洗,以便数据在训练集中和测试集中均匀分布。 尝试更多增强技术 (ref) 或尝试收集更多数据,因为现有数据集看起来非常小,这会导致过度拟合。 尝试将 dropout 从 0.67 降低到 0.2 左右。 添加减少过拟合的正则化技术。 使用earlystopping,当后续epoch的准确率没有太大提升时,可以调用earlystopping。【讨论】:
以上是关于TensorFlow 对象检测 API 过拟合的主要内容,如果未能解决你的问题,请参考以下文章