CasiaFan
commented
4 years ago Could you try to run model_main.py instead of trainer.py and see the results first? I'm not sure about the issue since it seems that the max_level is given 8. @Rublins
Open Rublins opened 4 years ago
CasiaFan
commented
4 years ago Could you try to run model_main.py instead of trainer.py and see the results first? I'm not sure about the issue since it seems that the max_level is given 8. @Rublins
Rublins
commented
4 years ago @CasiaFan.. Thank You, I solved the above issue.. Now i got Value error.. Kindly help me to sort out this issue.
python model_main.py --logtostderr --train_dir=Retina/Ret_Training --pipeline_config_path=Retina/Ret_Training/retinanet_50_train1.config
WARNING:tensorflow:Forced number of epochs for all eval validations to be 1.
WARNING:tensorflow:Expected number of evaluation epochs is 1, but instead encountered eval_on_train_input_config.num_epochs = 0. Overwriting num_epochs to 1.
WARNING:tensorflow:Using temporary folder as model directory: C:\Users\SATHIE~1\AppData\Local\Temp\tmp_pb1_uxg
WARNING:tensorflow:Estimator's model_fn (<function create_model_fn.tf.data.experimental.parallel_interleave(...).
WARNING:tensorflow:From C:\Users\Sathiesh Kumar\AppData\Local\Continuum\anaconda3\envs\retina\lib\site-packages\tensorflow\python\ops\sparse_ops.py:1165: sparse_to_dense (from tensorflow.python.ops.sparse_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Create a tf.sparse.SparseTensor and use tf.sparse.to_dense instead.
WARNING:tensorflow:From D:\RUBIN\RESEARCH\Tensorflow1\models-master\research\object_detection\builders\dataset_builder.py:158: batch_and_drop_remainder (from tensorflow.contrib.data.python.ops.batching) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.data.Dataset.batch(..., drop_remainder=True).
p2 shape: (4, 80, 80, 256)
p3 shape: (4, 40, 40, 512)
p4 shape: (4, 20, 20, 1024)
p5 shape: (4, 10, 10, 2048)
l3 shape: (4, 40, 40, 256)
l4 shape: (4, 20, 20, 256)
l5 shape: (4, 10, 10, 256)
p3 shape: (4, 40, 40, 256)
p4 shape: (4, 20, 20, 256)
Traceback (most recent call last):
File "model_main.py", line 109, in num_anchors_per_location.
Retinanet.py
import tensorflow as tf
import math
from object_detection.utils.shape_utils import combined_static_and_dynamic_shape
# tf.enable_eager_execution()
BN_PARAMS = {"bn_decay": 0.997,
"bn_epsilon": 1e-4}
# define number of layers of each block for different architecture
RESNET_ARCH_BLOCK = {"resnet50": [3, 4, 6, 3],
"resnet101": [3, 4, 23, 3]}
def nearest_neighbor_upsampling(input_tensor, scale):
"""Nearest neighbor upsampling implementation.
NOTE: See TensorFlow Object Detection API uitls.ops
Args:
input_tensor: A float32 tensor of size [batch, height_in, width_in, channels].
scale: An integer multiple to scale resolution of input data.
Returns:
upsample_input: A float32 tensor of size [batch, height_in*scale, width_in*scale, channels].
"""
with tf.name_scope('nearest_neighbor_upsampling'):
(batch_size, h, w, c) = combined_static_and_dynamic_shape(input_tensor)
output_tensor = tf.reshape(input_tensor, [batch_size, h, 1, w, 1, c]) * tf.ones(
[1, 1, scale, 1, scale, 1], dtype=input_tensor.dtype)
return tf.reshape(output_tensor, [batch_size, h*scale, w*scale, c])
def conv2d_same(inputs, depth, kernel_size, strides, scope=None):
with tf.name_scope(scope, None):
if strides == 1:
return tf.layers.conv2d(inputs, depth, kernel_size, padding='SAME')
else:
pad_total = kernel_size - 1
pad_beg = pad_total // 2
pad_end = pad_total - pad_beg
inputs = tf.pad(inputs, [[0, 0], [pad_beg, pad_end], [pad_beg, pad_end], [0, 0]])
return tf.layers.conv2d(inputs,
depth,
kernel_size,
strides=strides,
padding='VALID',
use_bias=False,
kernel_initializer=tf.variance_scaling_initializer())
def bn_with_relu(inputs, is_training, relu=True, init_zero=False, name=None):
if not init_zero:
gamma_init = tf.ones_initializer()
else:
gamma_init = tf.zeros_initializer()
inputs = tf.layers.batch_normalization(inputs,
training=is_training,
momentum=BN_PARAMS["bn_decay"],
epsilon=BN_PARAMS["bn_epsilon"],
scale=True,
fused=True,
gamma_initializer=gamma_init,
name=name)
if relu:
inputs = tf.nn.relu(inputs)
return inputs
def bottleneck(inputs, depth, strides, is_training, projection=False, scope=None):
"""Bottleneck residual unit variant with BN after convolutions
When putting together 2 consecutive ResNet blocks that use this unit,
one should use stride =2 in the last unit of first block
Args:
inputs: A tensor of size [batchsize, height, width, channels] (after BN)
depth: The depth of the block unit output
strides: the ResNet unit's stride. Determines the amount of downsampling of
the units output compared to its input
is_training: indicate training state for BN layer
projection: if this block will use a projection. True for first block in block groups
scope: Optional variable scope
Returns:
The ResNet unit output
"""
with tf.variable_scope(scope, 'bottleneck', [inputs]) as sc:
# shortcut connection
shortcut = inputs
depth_out = depth * 4
if projection:
shortcut = conv2d_same(shortcut, depth_out, kernel_size=1, strides=strides, scope='shortcut')
shortcut = bn_with_relu(shortcut, is_training, relu=False)
# layer1
residual = conv2d_same(inputs, depth, kernel_size=1, strides=1, scope='conv1')
residual = bn_with_relu(residual, is_training)
# layer 2
residual = conv2d_same(residual, depth, kernel_size=3, strides=strides, scope='conv2')
residual = bn_with_relu(residual, is_training)
# layer 3
residual = conv2d_same(residual, depth_out, kernel_size=1, strides=1, scope='conv3')
residual = bn_with_relu(residual, is_training, relu=False, init_zero=True)
output = shortcut + residual
return tf.nn.relu(output)
def stack_bottleneck(inputs, layers, depth, strides, is_training, scope=None):
""" Stack bottleneck planes
This function creates scopes for the ResNet in the form of 'block_name/plane_1, block_name/plane_2', etc.
Args:
layers: number of layers in this block
"""
with tf.variable_scope(scope, 'block', [inputs]) as sc:
inputs = bottleneck(inputs, depth, strides=strides, is_training=is_training, projection=True)
for i in range(1, layers):
layer_scope = "unit_{}".format(i)
inputs = bottleneck(inputs, depth, strides=1, is_training=is_training, scope=layer_scope)
return inputs
def retinanet_fpn(inputs,
block_layers,
depth=256,
is_training=True,
scope=None):
"""
Generator for RetinaNet FPN models. A small modification of initial FPN model for returning layers
{P3, P4, P5, P6, P7}. See paper Focal Loss for Dense Object Detection. arxiv: 1708.02002
P2 is discarded and P6 is obtained via 3x3 stride-2 conv on c5; P7 is computed by applying ReLU followed by
3x3 stride-2 conv on P6. P7 is to improve large object detection
Returns:
5 feature map tensors: {P3, P4, P5, P6, P7}
"""
with tf.variable_scope(scope, 'retinanet_fpn', [inputs]) as sc:
net = conv2d_same(inputs, 64, kernel_size=7, strides=2, scope='conv1')
net = bn_with_relu(net, is_training)
net = tf.layers.max_pooling2d(net, pool_size=3, strides=2, padding='SAME', name='pool1')
# Bottom up
# block 1, down-sampling is done in conv3_1, conv4_1, conv5_1
p2 = stack_bottleneck(net, layers=block_layers[0], depth=64, strides=1, is_training=is_training)
# block 2
p3 = stack_bottleneck(p2, layers=block_layers[1], depth=128, strides=2, is_training=is_training)
# block 3
p4 = stack_bottleneck(p3, layers=block_layers[2], depth=256, strides=2, is_training=is_training)
# block 4
p5 = stack_bottleneck(p4, layers=block_layers[3], depth=512, strides=2, is_training=is_training)
# lateral layer
l3 = tf.layers.conv2d(p3, filters=depth, kernel_size=1, strides=1, name='l3', padding='SAME')
l4 = tf.layers.conv2d(p4, filters=depth, kernel_size=1, strides=1, name='l4', padding='SAME')
l5 = tf.layers.conv2d(p5, filters=depth, kernel_size=1, strides=1, name='l5', padding='SAME')
print ('p2 shape:', p2.get_shape())
print ('p3 shape:', p3.get_shape())
print ('p4 shape:', p4.get_shape())
print ('p5 shape:', p5.get_shape())
print ('l3 shape:', l3.get_shape())
print ('l4 shape:', l4.get_shape())
print ('l5 shape:', l5.get_shape())
# Top down
p4 = nearest_neighbor_upsampling(l5, 2) + l4
p3 = nearest_neighbor_upsampling(p4, 2) + l3
print ('p3 shape:', p3.get_shape())
print ('p4 shape:', p4.get_shape())
# add post-hoc conv layers
p3 = tf.layers.conv2d(p3, filters=depth, kernel_size=3, strides=1, padding='SAME', name='post-hoc-d3')
p4 = tf.layers.conv2d(p4, filters=depth, kernel_size=3, strides=1, padding='SAME', name='post-hoc-d4')
p5 = tf.layers.conv2d(l5, filters=depth, kernel_size=3, strides=1, padding='SAME', name='post-hoc-d5')
# coarse layer: 6, 7
# p6
p6 = tf.layers.conv2d(p5, filters=depth, kernel_size=3, strides=2, name='conv6', padding='SAME')
p6 = tf.nn.relu(p6)
# P7
p7 = tf.layers.conv2d(p6, filters=depth, kernel_size=3, strides=2, name='conv7', padding='SAME')
# add normalization to each layer
features = {3: p3,
4: p4,
5: l5,
6: p6,
7: p7}
for layer in features:
features[layer] = tf.layers.batch_normalization(features[layer],
training=is_training,
momentum=BN_PARAMS["bn_decay"],
epsilon=BN_PARAMS["bn_epsilon"],
center=True,
scale=True,
fused=True,
name='p{}-bn'.format(layer))
return features
def share_weight_class_net(inputs, level, num_classes, num_anchors_per_loc, num_layers_before_predictor=4, is_training=True):
"""
net for predicting class labels
NOTE: Share same weights when called more then once on different feature maps
Args:
inputs: feature map with shape (batch_size, h, w, channel)
level: which feature map
num_classes: number of predicted classes
num_anchors_per_loc: number of anchors at each spatial location in feature map
num_layers_before_predictor: number of the additional conv layers before the predictor.
is_training: is in training or not
returns:
feature with shape (batch_size, h, w, num_classes*num_anchors)
"""
for i in range(num_layers_before_predictor):
inputs = tf.layers.conv2d(inputs, filters=256, kernel_size=3, strides=1,
kernel_initializer=tf.random_normal_initializer(stddev=0.01),
bias_initializer=tf.zeros_initializer(),
padding="SAME",
name='class_{}'.format(i))
inputs = bn_with_relu(inputs, is_training, relu=True, init_zero=False, name="class_{}_bn_level_{}".format(i, level))
outputs = tf.layers.conv2d(inputs,
filters=num_classes*num_anchors_per_loc,
kernel_size=3,
bias_initializer=tf.constant_initializer(-math.log((1 - 0.01) / 0.01)),
kernel_initializer=tf.random_normal_initializer(stddev=0.01),
padding="SAME",
name="class_pred")
return outputs
def share_weight_box_net(inputs, level, num_anchors_per_loc, num_layers_before_predictor=4, is_training=True):
"""
Similar to class_net with output feature shape (batch_size, h, w, num_anchors*4)
"""
for i in range(num_layers_before_predictor):
inputs = tf.layers.conv2d(inputs, filters=256, kernel_size=3, strides=1,
bias_initializer=tf.zeros_initializer(),
kernel_initializer=tf.random_normal_initializer(stddev=0.01),
padding="SAME",
name='box_{}'.format(i))
inputs = bn_with_relu(inputs, is_training, relu=True, init_zero=False, name="box_{}_bn_level_{}".format(i, level))
outputs = tf.layers.conv2d(inputs,
filters=4*num_anchors_per_loc,
kernel_size=3,
kernel_initializer=tf.random_normal_initializer(stddev=0.01),
padding="SAME",
name="box_pred")
return outputs
def retinanet(images, num_classes, num_anchors_per_loc, resnet_arch='resnet50', is_training=True):
"""
Get box prediction features and class prediction features from given images
Args:
images: input batch of images with shape (batch_size, h, w, 3)
num_classes: number of classes for prediction
num_anchors_per_loc: number of anchors at each feature map spatial location
resnet_arch: name of which resnet architecture used
is_training: indicate training or not
return:
prediciton dict: holding following items:
box_predictions tensor from each feature map with shape (batch_size, num_anchors, 4)
class_predictions_with_bg tensor from each feature map with shape (batch_size, num_anchors, num_class+1)
feature_maps: list of tensor of feature map
"""
assert resnet_arch in list(RESNET_ARCH_BLOCK.keys()), "resnet architecture not defined"
with tf.variable_scope('retinanet'):
batch_size = combined_static_and_dynamic_shape(images)[0]
features = retinanet_fpn(images, block_layers=RESNET_ARCH_BLOCK[resnet_arch], is_training=is_training)
class_pred = []
box_pred = []
feature_map_list = []
num_slots = num_classes + 1
with tf.variable_scope('class_net', reuse=tf.AUTO_REUSE):
for level in features.keys():
class_outputs = share_weight_class_net(features[level], level,
num_slots,
num_anchors_per_loc,
is_training=is_training)
class_outputs = tf.reshape(class_outputs, shape=[batch_size, -1, num_slots])
class_pred.append(class_outputs)
feature_map_list.append(features[level])
with tf.variable_scope('box_net', reuse=tf.AUTO_REUSE):
for level in features.keys():
box_outputs = share_weight_box_net(features[level], level, num_anchors_per_loc, is_training=is_training)
box_outputs = tf.reshape(box_outputs, shape=[batch_size, -1, 4])
box_pred.append(box_outputs)
return get(box_pred=tf.concat(box_pred, axis=1),
cls_pred=tf.concat(class_pred, axis=1),
feature_map_list=feature_map_list)Config file
# SSD with Resnet 50 v1 FPN feature extractor, shared box predictor and focal
# loss (a.k.a Retinanet).
# See Lin et al, https://arxiv.org/abs/1708.02002
# Trained on COCO, initialized from Imagenet classification checkpoint
# Achieves 35.2 mAP on COCO14 minival dataset. Doubling the number of training
# steps to 50k gets 36.9 mAP
# This config is TPU compatible
model {
ssd {
inplace_batchnorm_update: true
freeze_batchnorm: false
num_classes: 10
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: 2
}
}
image_resizer {
fixed_shape_resizer {
height: 640
width: 640
}
}
box_predictor {
weight_shared_convolutional_box_predictor {
depth: 256
class_prediction_bias_init: -4.6
conv_hyperparams {
activation: RELU_6,
regularizer {
l2_regularizer {
weight: 0.0004
}
}
initializer {
random_normal_initializer {
stddev: 0.01
mean: 0.0
}
}
batch_norm {
scale: true,
decay: 0.997,
epsilon: 0.001,
}
}
num_layers_before_predictor: 4
kernel_size: 3
}
}
feature_extractor {
type: 'retinanet_50'
min_depth: 0
depth_multiplier: 1.0
pad_to_multiple: 4
override_base_feature_extractor_hyperparams: true
}
loss {
classification_loss {
weighted_sigmoid_focal {
alpha: 0.25
gamma: 2.0
}
}
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: 0.6
max_detections_per_class: 100
max_total_detections: 100
}
score_converter: SIGMOID
}
}
}
train_config: {
#fine_tune_checkpoint: "D:/RUBIN/RESEARCH/Tensorflow1/models/research/object_detection/ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03/model.ckpt"
batch_size: 4
sync_replicas: false
startup_delay_steps: 0
replicas_to_aggregate: 2
num_steps: 5000
data_augmentation_options {
random_horizontal_flip {
}
}
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
}
}
optimizer {
momentum_optimizer: {
learning_rate: {
cosine_decay_learning_rate {
learning_rate_base: .04
total_steps: 5000
warmup_learning_rate: .013333
warmup_steps: 2000
}
}
momentum_optimizer_value: 0.9
}
use_moving_average: false
}
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
}
train_input_reader: {
tf_record_input_reader {
input_path: "D:/RUBIN/RESEARCH/Tensorflow1/models/research/object_detection/ORI_Image_300/train1.record"
}
label_map_path: "D:/RUBIN/RESEARCH/Tensorflow1/models-master/research/object_detection/Retina/Ret_Training/labelmap.pbtxt"
}
eval_config: {
metrics_set: "coco_detection_metrics"
num_examples: 1424
num_visualizations:1424
max_evals: 1
}
eval_input_reader: {
tf_record_input_reader {
input_path: "D:/RUBIN/RESEARCH/Tensorflow1/models/research/object_detection/ORI_Image_300/test1.record"
}
label_map_path: "D:/RUBIN/RESEARCH/Tensorflow1/models-master/research/object_detection/Retina/Ret_Training/labelmap.pbtxt"
shuffle: false
num_readers: 1
}Thank you.
CasiaFan
commented
4 years ago I test this code using tf 1.14 but things run okay. This error should a mismatch between the number of feature maps (5 layers) and the number of anchor grid settings (which is max_level-min_level+1). So first, could you check the tf model api version? Anyway, I find your input image size is 640, but in that case, the p2 layer shape should be 160x160 or 1/4 of the input width and height while your result is 80. I am not sure where is the cause of this problem.
@Rublins
Rublins
commented
4 years ago Thank you @CasiaFan... I changed the input size to 320 x 320 and now it works fine for ResNet-50..
Rublins
commented
4 years ago @CasiaFan can you suggest any suitable pre-trained model weights for RetinaNet-101 (ResNet-101 as a feature extractor). Thank You
CasiaFan
commented
4 years ago @Rublins If you are using my custom code for training, I'm afraid there would be no pre-trained resnet101 weights due to the node name mismatch. But if you are using the official TF object detection api, this model weights trained on Open Images Dataset should meet your need.
Help me to clear this KeyError :8
Traceback (most recent call last): File "train.py", line 184, in
tf.app.run()
File "C:\Users\Sathiesh Kumar\AppData\Local\Continuum\anaconda3\envs\retina\lib\site-packages\tensorflow\python\platform\app.py", line 125, in run
_sys.exit(main(argv))
File "C:\Users\Sathiesh Kumar\AppData\Local\Continuum\anaconda3\envs\retina\lib\site-packages\tensorflow\python\util\deprecation.py", line 272, in new_func
return func(*args, *kwargs)
File "train.py", line 180, in main
graph_hook_fn=graph_rewriter_fn)
File "D:\RUBIN\RESEARCH\Tensorflow1\models-master\research\object_detection\legacy\trainer.py", line 292, in train
clones = model_deploy.create_clones(deploy_config, model_fn, [input_queue])
File "D:\RUBIN\RESEARCH\Tensorflow1\models-master\research\slim\deployment\model_deploy.py", line 194, in create_clones
outputs = model_fn(args, **kwargs)
File "D:\RUBIN\RESEARCH\Tensorflow1\models-master\research\object_detection\legacy\trainer.py", line 205, in _create_losses
prediction_dict = detection_model.predict(images, true_image_shapes)
File "D:\RUBIN\RESEARCH\Tensorflow1\models-master\research\object_detection\meta_architectures\ssd_meta_arch.py", line 600, in predict
preprocessed_inputs)
File "D:\RUBIN\RESEARCH\Tensorflow1\models-master\research\object_detection\models\retinanet_feature_extractor.py", line 135, in extract_features
return [image_features[x] for x in range(self._min_level, self._max_level+1)]
File "D:\RUBIN\RESEARCH\Tensorflow1\models-master\research\object_detection\models\retinanet_feature_extractor.py", line 135, in
return [image_features[x] for x in range(self._min_level, self._max_level+1)]
KeyError: 8
I have attached the config file used for training.
SSD with Resnet 50 v1 FPN feature extractor, shared box predictor and focal
loss (a.k.a Retinanet).
See Lin et al, https://arxiv.org/abs/1708.02002
Trained on COCO, initialized from Imagenet classification checkpoint
Achieves 35.2 mAP on COCO14 minival dataset. Doubling the number of training
steps to 50k gets 36.9 mAP
This config is TPU compatible
model { ssd { inplace_batchnorm_update: true freeze_batchnorm: false num_classes: 10 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: 2 } } image_resizer { fixed_shape_resizer { height: 640 width: 640 } } box_predictor { weight_shared_convolutional_box_predictor { depth: 256 class_prediction_bias_init: -4.6 conv_hyperparams { activation: RELU_6, regularizer { l2_regularizer { weight: 0.0004 } } initializer { random_normal_initializer { stddev: 0.01 mean: 0.0 } } batch_norm { scale: true, decay: 0.997, epsilon: 0.001, } } num_layers_before_predictor: 4 kernel_size: 3 } } feature_extractor { type: 'retinanet_50' min_depth: 0 depth_multiplier: 1.0 pad_to_multiple: 32 override_base_feature_extractor_hyperparams: true } loss { classification_loss { weighted_sigmoid_focal { alpha: 0.25 gamma: 2.0 } } 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: 0.6 max_detections_per_class: 100 max_total_detections: 100 } score_converter: SIGMOID } } }
train_config: {
fine_tune_checkpoint: "D:/RUBIN/RESEARCH/Tensorflow1/models/research/object_detection/ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03/model.ckpt"
batch_size: 1 sync_replicas: false startup_delay_steps: 0 replicas_to_aggregate: 2 num_steps: 5000 data_augmentation_options { random_horizontal_flip { } } 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 } } optimizer { momentum_optimizer: { learning_rate: { cosine_decay_learning_rate { learning_rate_base: .04 total_steps: 5000 warmup_learning_rate: .013333 warmup_steps: 2000 } } momentum_optimizer_value: 0.9 } use_moving_average: false } max_number_of_boxes: 100 unpad_groundtruth_tensors: false }
train_input_reader: { tf_record_input_reader { input_path: "D:/RUBIN/RESEARCH/Tensorflow1/models/research/object_detection/ORI_Image_300/train1.record" } label_map_path: "D:/RUBIN/RESEARCH/Tensorflow1/models-master/research/object_detection/Retina/Ret_Training/labelmap.pbtxt" }
eval_config: { metrics_set: "coco_detection_metrics" num_examples: 1424 num_visualizations:1424 max_evals: 1 }
eval_input_reader: { tf_record_input_reader { input_path: "D:/RUBIN/RESEARCH/Tensorflow1/models/research/object_detection/ORI_Image_300/test1.record" } label_map_path: "D:/RUBIN/RESEARCH/Tensorflow1/models-master/research/object_detection/Retina/Ret_Training/labelmap.pbtxt" shuffle: false num_readers: 1 }
Thank You.