东方耀AI技术分享

 找回密码
 立即注册

QQ登录

只需一步,快速开始

搜索
热搜: 活动 交友 discuz
查看: 3370|回复: 2

[课堂笔记] python2.7.15下跑通caffe-ssd模型训练widerface人脸数据集

[复制链接]

1365

主题

1856

帖子

1万

积分

管理员

Rank: 10Rank: 10Rank: 10

积分
14418
QQ
发表于 2019-8-6 08:37:41 | 显示全部楼层 |阅读模式


python2.7.15下跑通caffe-ssd模型训练widerface人脸数据集

下面的代码是从examples/ssd/ssd_pascal.py拷贝的,自己修改的地方用红色标示

# -*- coding: utf-8 -*-
from __future__ import print_function
__author__ = u'东方耀 微信:dfy_88888'
__date__ = '2019/8/6 上午8:33'
__product__ = 'PyCharm'
__filename__ = 'ssd_face_dfy'


import caffe
from caffe.model_libs import *
from google.protobuf import text_format

import math
import os
import shutil
import stat
import subprocess
import sys


# Add extra layers on top of a "base" network (e.g. VGGNet or Inception).
def AddExtraLayers(net, use_batchnorm=True, lr_mult=1):
    use_relu = True

    # Add additional convolutional layers.
    # 19 x 19
    from_layer = net.keys()[-1]

    # TODO(weiliu89): Construct the name using the last layer to avoid duplication.
    # 10 x 10
    out_layer = "conv6_1"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 256, 1, 0, 1,
                lr_mult=lr_mult)

    from_layer = out_layer
    out_layer = "conv6_2"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 512, 3, 1, 2,
                lr_mult=lr_mult)

    # 5 x 5
    from_layer = out_layer
    out_layer = "conv7_1"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 128, 1, 0, 1,
                lr_mult=lr_mult)

    from_layer = out_layer
    out_layer = "conv7_2"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 256, 3, 1, 2,
                lr_mult=lr_mult)

    # 3 x 3
    from_layer = out_layer
    out_layer = "conv8_1"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 128, 1, 0, 1,
                lr_mult=lr_mult)

    from_layer = out_layer
    out_layer = "conv8_2"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 256, 3, 0, 1,
                lr_mult=lr_mult)

    # 1 x 1
    from_layer = out_layer
    out_layer = "conv9_1"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 128, 1, 0, 1,
                lr_mult=lr_mult)

    from_layer = out_layer
    out_layer = "conv9_2"
    ConvBNLayer(net, from_layer, out_layer, use_batchnorm, use_relu, 256, 3, 0, 1,
                lr_mult=lr_mult)

    return net


### Modify the following parameters accordingly ###
# The directory which contains the caffe code.
# We assume you are running the script at the CAFFE_ROOT.
caffe_root = os.getcwd()

# Set true if you want to start training right after generating all files.
run_soon = True
# Set true if you want to load from most recently saved snapshot.
# Otherwise, we will load from the pretrain_model defined below.
resume_training = True
# If true, Remove old model files.
remove_old_models = False

# The database file for training data. Created by data/VOC0712/create_data.sh
train_data = "/home/dfy888/DataSets/数据集LMDB/WiderFace/lmdb/WiderFace_trainval_lmdb"
# The database file for testing data. Created by data/VOC0712/create_data.sh
test_data = "/home/dfy888/DataSets/数据集LMDB/WiderFace/lmdb/WiderFace_test_lmdb"
# Specify the batch sampler.
resize_width = 300
resize_height = 300
resize = "{}x{}".format(resize_width, resize_height)
batch_sampler = [
    {
        'sampler': {
        },
        'max_trials': 1,
        'max_sample': 1,
    },
    {
        'sampler': {
            'min_scale': 0.3,
            'max_scale': 1.0,
            'min_aspect_ratio': 0.5,
            'max_aspect_ratio': 2.0,
        },
        'sample_constraint': {
            'min_jaccard_overlap': 0.1,
        },
        'max_trials': 50,
        'max_sample': 1,
    },
    {
        'sampler': {
            'min_scale': 0.3,
            'max_scale': 1.0,
            'min_aspect_ratio': 0.5,
            'max_aspect_ratio': 2.0,
        },
        'sample_constraint': {
            'min_jaccard_overlap': 0.3,
        },
        'max_trials': 50,
        'max_sample': 1,
    },
    {
        'sampler': {
            'min_scale': 0.3,
            'max_scale': 1.0,
            'min_aspect_ratio': 0.5,
            'max_aspect_ratio': 2.0,
        },
        'sample_constraint': {
            'min_jaccard_overlap': 0.5,
        },
        'max_trials': 50,
        'max_sample': 1,
    },
    {
        'sampler': {
            'min_scale': 0.3,
            'max_scale': 1.0,
            'min_aspect_ratio': 0.5,
            'max_aspect_ratio': 2.0,
        },
        'sample_constraint': {
            'min_jaccard_overlap': 0.7,
        },
        'max_trials': 50,
        'max_sample': 1,
    },
    {
        'sampler': {
            'min_scale': 0.3,
            'max_scale': 1.0,
            'min_aspect_ratio': 0.5,
            'max_aspect_ratio': 2.0,
        },
        'sample_constraint': {
            'min_jaccard_overlap': 0.9,
        },
        'max_trials': 50,
        'max_sample': 1,
    },
    {
        'sampler': {
            'min_scale': 0.3,
            'max_scale': 1.0,
            'min_aspect_ratio': 0.5,
            'max_aspect_ratio': 2.0,
        },
        'sample_constraint': {
            'max_jaccard_overlap': 1.0,
        },
        'max_trials': 50,
        'max_sample': 1,
    },
]
train_transform_param = {
    'mirror': True,
    'mean_value': [104, 117, 123],
    'resize_param': {
        'prob': 1,
        'resize_mode': P.Resize.WARP,
        'height': resize_height,
        'width': resize_width,
        'interp_mode': [
            P.Resize.LINEAR,
            P.Resize.AREA,
            P.Resize.NEAREST,
            P.Resize.CUBIC,
            P.Resize.LANCZOS4,
        ],
    },
    'distort_param': {
        'brightness_prob': 0.5,
        'brightness_delta': 32,
        'contrast_prob': 0.5,
        'contrast_lower': 0.5,
        'contrast_upper': 1.5,
        'hue_prob': 0.5,
        'hue_delta': 18,
        'saturation_prob': 0.5,
        'saturation_lower': 0.5,
        'saturation_upper': 1.5,
        'random_order_prob': 0.0,
    },
    'expand_param': {
        'prob': 0.5,
        'max_expand_ratio': 4.0,
    },
    'emit_constraint': {
        'emit_type': caffe_pb2.EmitConstraint.CENTER,
    }
}
test_transform_param = {
    'mean_value': [104, 117, 123],
    'resize_param': {
        'prob': 1,
        'resize_mode': P.Resize.WARP,
        'height': resize_height,
        'width': resize_width,
        'interp_mode': [P.Resize.LINEAR],
    },
}

# If true, use batch norm for all newly added layers.
# Currently only the non batch norm version has been tested.
use_batchnorm = False
lr_mult = 1
# Use different initial learning rate.
if use_batchnorm:
    base_lr = 0.0004
else:
    # A learning rate for batch_size = 1, num_gpus = 1.
    base_lr = 0.00004

# Modify the job name if you want.
job_name = "Caffe_SSD_{}".format(resize)
# The name of the model. Modify it if you want.
model_name = "VGG_WiderFace_{}".format(job_name)

# Directory which stores the model .prototxt file.
save_dir = "models/VGGNet/WiderFace/{}".format(job_name)
# Directory which stores the snapshot of models.
snapshot_dir = "models/VGGNet/WiderFace/{}".format(job_name)
# Directory which stores the job script and log file.
job_dir = "jobs/VGGNet/WiderFace/{}".format(job_name)
# Directory which stores the detection results.
output_result_dir = "{}/data/VOCdevkit/results/WiderFace/{}/Main".format(os.environ['HOME'], job_name)


# model definition files.
train_net_file = "{}/train.prototxt".format(save_dir)
test_net_file = "{}/test.prototxt".format(save_dir)
deploy_net_file = "{}/deploy.prototxt".format(save_dir)
solver_file = "{}/solver.prototxt".format(save_dir)
# snapshot prefix.
snapshot_prefix = "{}/{}".format(snapshot_dir, model_name)
# job script path.
job_file = "{}/{}.sh".format(job_dir, model_name)

# Stores the test image names and sizes. Created by data/VOC0712/create_list.sh
name_size_file = "/home/dfy888/py2_caffe_source_works/caffe-ssd-py2/data/WiderFace/test_name_size.txt"
# The pretrained model. We use the Fully convolutional reduced (atrous) VGGNet.
pretrain_model = ""
# Stores LabelMapItem.
label_map_file = "/home/dfy888/py2_caffe_source_works/caffe-ssd-py2/data/WiderFace/labelmap_voc.prototxt"

# MultiBoxLoss parameters.
num_classes = 2
share_location = True
background_label_id = 0
train_on_diff_gt = True
normalization_mode = P.Loss.VALID
code_type = P.PriorBox.CENTER_SIZE
ignore_cross_boundary_bbox = False
mining_type = P.MultiBoxLoss.MAX_NEGATIVE
neg_pos_ratio = 3.
loc_weight = (neg_pos_ratio + 1.) / 4.
multibox_loss_param = {
    'loc_loss_type': P.MultiBoxLoss.SMOOTH_L1,
    'conf_loss_type': P.MultiBoxLoss.SOFTMAX,
    'loc_weight': loc_weight,
    'num_classes': num_classes,
    'share_location': share_location,
    'match_type': P.MultiBoxLoss.PER_PREDICTION,
    'overlap_threshold': 0.5,
    'use_prior_for_matching': True,
    'background_label_id': background_label_id,
    'use_difficult_gt': train_on_diff_gt,
    'mining_type': mining_type,
    'neg_pos_ratio': neg_pos_ratio,
    'neg_overlap': 0.5,
    'code_type': code_type,
    'ignore_cross_boundary_bbox': ignore_cross_boundary_bbox,
}
loss_param = {
    'normalization': normalization_mode,
}

# parameters for generating priors.
# minimum dimension of input image
min_dim = 300
# conv4_3 ==> 38 x 38
# fc7 ==> 19 x 19
# conv6_2 ==> 10 x 10
# conv7_2 ==> 5 x 5
# conv8_2 ==> 3 x 3
# conv9_2 ==> 1 x 1
mbox_source_layers = ['conv4_3', 'fc7', 'conv6_2', 'conv7_2', 'conv8_2', 'conv9_2']
# in percent %
min_ratio = 20
max_ratio = 90
step = int(math.floor((max_ratio - min_ratio) / (len(mbox_source_layers) - 2)))
min_sizes = []
max_sizes = []
for ratio in xrange(min_ratio, max_ratio + 1, step):
    min_sizes.append(min_dim * ratio / 100.)
    max_sizes.append(min_dim * (ratio + step) / 100.)
min_sizes = [min_dim * 10 / 100.] + min_sizes
max_sizes = [min_dim * 20 / 100.] + max_sizes
steps = [8, 16, 32, 64, 100, 300]
aspect_ratios = [[2], [2, 3], [2, 3], [2, 3], [2], [2]]
# L2 normalize conv4_3.
normalizations = [20, -1, -1, -1, -1, -1]
# variance used to encode/decode prior bboxes.
if code_type == P.PriorBox.CENTER_SIZE:
    prior_variance = [0.1, 0.1, 0.2, 0.2]
else:
    prior_variance = [0.1]
flip = True
clip = False

# Solver parameters.
# Defining which GPUs to use.
gpus = "0"
gpulist = gpus.split(",")
num_gpus = len(gpulist)

# Divide the mini-batch to different GPUs.
batch_size = 32
accum_batch_size = 32
iter_size = accum_batch_size / batch_size
solver_mode = P.Solver.CPU
device_id = 0
batch_size_per_device = batch_size
if num_gpus > 0:
    batch_size_per_device = int(math.ceil(float(batch_size) / num_gpus))
    iter_size = int(math.ceil(float(accum_batch_size) / (batch_size_per_device * num_gpus)))
    solver_mode = P.Solver.GPU
    device_id = int(gpulist[0])

if normalization_mode == P.Loss.NONE:
    base_lr /= batch_size_per_device
elif normalization_mode == P.Loss.VALID:
    base_lr *= 25. / loc_weight
elif normalization_mode == P.Loss.FULL:
    # Roughly there are 2000 prior bboxes per image.
    # TODO(weiliu89): Estimate the exact # of priors.
    base_lr *= 2000.

# Evaluate on whole test set.
num_test_image = 1000
test_batch_size = 8
# Ideally test_batch_size should be divisible by num_test_image,
# otherwise mAP will be slightly off the true value.
test_iter = int(math.ceil(float(num_test_image) / test_batch_size))

solver_param = {
    # Train parameters
    'base_lr': base_lr,
    'weight_decay': 0.0005,
    'lr_policy': "multistep",
    'stepvalue': [80000, 100000, 120000],
    'gamma': 0.1,
    'momentum': 0.9,
    'iter_size': iter_size,
    'max_iter': 120000,
    'snapshot': 80000,
    'display': 10,
    'average_loss': 10,
    'type': "SGD",
    'solver_mode': solver_mode,
    'device_id': device_id,
    'debug_info': False,
    'snapshot_after_train': True,
    # Test parameters
    'test_iter': [test_iter],
    'test_interval': 10000,
    'eval_type': "detection",
    'ap_version': "11point",
    'test_initialization': False,
}

# parameters for generating detection output.
det_out_param = {
    'num_classes': num_classes,
    'share_location': share_location,
    'background_label_id': background_label_id,
    'nms_param': {'nms_threshold': 0.45, 'top_k': 400},
    'save_output_param': {
        'output_directory': output_result_dir,
        'output_name_prefix': "comp4_det_test_",
        'output_format': "VOC",
        'label_map_file': label_map_file,
        'name_size_file': name_size_file,
        'num_test_image': num_test_image,
    },
    'keep_top_k': 200,
    'confidence_threshold': 0.01,
    'code_type': code_type,
}

# parameters for evaluating detection results.
det_eval_param = {
    'num_classes': num_classes,
    'background_label_id': background_label_id,
    'overlap_threshold': 0.5,
    'evaluate_difficult_gt': False,
    'name_size_file': name_size_file,
}

### Hopefully you don't need to change the following ###
# Check file.
check_if_exist(train_data)
check_if_exist(test_data)
check_if_exist(label_map_file)
check_if_exist(pretrain_model)
make_if_not_exist(save_dir)
make_if_not_exist(job_dir)
make_if_not_exist(snapshot_dir)

# Create train net.
net = caffe.NetSpec()
net.data, net.label = CreateAnnotatedDataLayer(train_data, batch_size=batch_size_per_device,
                                               train=True, output_label=True, label_map_file=label_map_file,
                                               transform_param=train_transform_param, batch_sampler=batch_sampler)

VGGNetBody(net, from_layer='data', fully_conv=True, reduced=True, dilated=True,
           dropout=False)

AddExtraLayers(net, use_batchnorm, lr_mult=lr_mult)

mbox_layers = CreateMultiBoxHead(net, data_layer='data', from_layers=mbox_source_layers,
                                 use_batchnorm=use_batchnorm, min_sizes=min_sizes, max_sizes=max_sizes,
                                 aspect_ratios=aspect_ratios, steps=steps, normalizations=normalizations,
                                 num_classes=num_classes, share_location=share_location, flip=flip, clip=clip,
                                 prior_variance=prior_variance, kernel_size=3, pad=1, lr_mult=lr_mult)

# Create the MultiBoxLossLayer.
name = "mbox_loss"
mbox_layers.append(net.label)
net[name] = L.MultiBoxLoss(*mbox_layers, multibox_loss_param=multibox_loss_param,
                           loss_param=loss_param, include=dict(phase=caffe_pb2.Phase.Value('TRAIN')),
                           propagate_down=[True, True, False, False])

with open(train_net_file, 'w') as f:
    print('name: "{}_train"'.format(model_name), file=f)
    print(net.to_proto(), file=f)
shutil.copy(train_net_file, job_dir)

# Create test net.
net = caffe.NetSpec()
net.data, net.label = CreateAnnotatedDataLayer(test_data, batch_size=test_batch_size,
                                               train=False, output_label=True, label_map_file=label_map_file,
                                               transform_param=test_transform_param)

VGGNetBody(net, from_layer='data', fully_conv=True, reduced=True, dilated=True,
           dropout=False)

AddExtraLayers(net, use_batchnorm, lr_mult=lr_mult)

mbox_layers = CreateMultiBoxHead(net, data_layer='data', from_layers=mbox_source_layers,
                                 use_batchnorm=use_batchnorm, min_sizes=min_sizes, max_sizes=max_sizes,
                                 aspect_ratios=aspect_ratios, steps=steps, normalizations=normalizations,
                                 num_classes=num_classes, share_location=share_location, flip=flip, clip=clip,
                                 prior_variance=prior_variance, kernel_size=3, pad=1, lr_mult=lr_mult)

conf_name = "mbox_conf"
if multibox_loss_param["conf_loss_type"] == P.MultiBoxLoss.SOFTMAX:
    reshape_name = "{}_reshape".format(conf_name)
    net[reshape_name] = L.Reshape(net[conf_name], shape=dict(dim=[0, -1, num_classes]))
    softmax_name = "{}_softmax".format(conf_name)
    net[softmax_name] = L.Softmax(net[reshape_name], axis=2)
    flatten_name = "{}_flatten".format(conf_name)
    net[flatten_name] = L.Flatten(net[softmax_name], axis=1)
    mbox_layers[1] = net[flatten_name]
elif multibox_loss_param["conf_loss_type"] == P.MultiBoxLoss.LOGISTIC:
    sigmoid_name = "{}_sigmoid".format(conf_name)
    net[sigmoid_name] = L.Sigmoid(net[conf_name])
    mbox_layers[1] = net[sigmoid_name]

net.detection_out = L.DetectionOutput(*mbox_layers,
                                      detection_output_param=det_out_param,
                                      include=dict(phase=caffe_pb2.Phase.Value('TEST')))
net.detection_eval = L.DetectionEvaluate(net.detection_out, net.label,
                                         detection_evaluate_param=det_eval_param,
                                         include=dict(phase=caffe_pb2.Phase.Value('TEST')))

with open(test_net_file, 'w') as f:
    print('name: "{}_test"'.format(model_name), file=f)
    print(net.to_proto(), file=f)
shutil.copy(test_net_file, job_dir)

# Create deploy net.
# Remove the first and last layer from test net.
deploy_net = net
with open(deploy_net_file, 'w') as f:
    net_param = deploy_net.to_proto()
    # Remove the first (AnnotatedData) and last (DetectionEvaluate) layer from test net.
    del net_param.layer[0]
    del net_param.layer[-1]
    net_param.name = '{}_deploy'.format(model_name)
    net_param.input.extend(['data'])
    net_param.input_shape.extend([
        caffe_pb2.BlobShape(dim=[1, 3, resize_height, resize_width])])
    print(net_param, file=f)
shutil.copy(deploy_net_file, job_dir)

# Create solver.
solver = caffe_pb2.SolverParameter(
    train_net=train_net_file,
    test_net=[test_net_file],
    snapshot_prefix=snapshot_prefix,
    **solver_param)

with open(solver_file, 'w') as f:
    print(solver, file=f)
shutil.copy(solver_file, job_dir)

max_iter = 0
# Find most recent snapshot.
for file in os.listdir(snapshot_dir):
    if file.endswith(".solverstate"):
        basename = os.path.splitext(file)[0]
        iter = int(basename.split("{}_iter_".format(model_name))[1])
        if iter > max_iter:
            max_iter = iter

# train_src_param = '--weights="{}" \\\n'.format(pretrain_model)
# 预训练模型为空 不使用
train_src_param = ''
if resume_training:
    if max_iter > 0:
        train_src_param = '--snapshot="{}_iter_{}.solverstate" \\\n'.format(snapshot_prefix, max_iter)

if remove_old_models:
    # Remove any snapshots smaller than max_iter.
    for file in os.listdir(snapshot_dir):
        if file.endswith(".solverstate"):
            basename = os.path.splitext(file)[0]
            iter = int(basename.split("{}_iter_".format(model_name))[1])
            if max_iter > iter:
                os.remove("{}/{}".format(snapshot_dir, file))
        if file.endswith(".caffemodel"):
            basename = os.path.splitext(file)[0]
            iter = int(basename.split("{}_iter_".format(model_name))[1])
            if max_iter > iter:
                os.remove("{}/{}".format(snapshot_dir, file))

# Create job file.
with open(job_file, 'w') as f:
    f.write('cd {}\n'.format(caffe_root))
    f.write('./build/tools/caffe train \\\n')
    f.write('--solver="{}" \\\n'.format(solver_file))
    f.write(train_src_param)
    if solver_param['solver_mode'] == P.Solver.GPU:
        f.write('--gpu {} 2>&1 | tee {}/{}.log\n'.format(gpus, job_dir, model_name))
    else:
        f.write('2>&1 | tee {}/{}.log\n'.format(job_dir, model_name))

# Copy the python script to job_dir.
py_file = os.path.abspath(__file__)
shutil.copy(py_file, job_dir)

# Run the job.
os.chmod(job_file, stat.S_IRWXU)
if run_soon:
    subprocess.call(job_file, shell=True)







在进入Caffe-SSD的训练环节时,总是出现如下报错:


math_functions.cpp:250] Check failed: a <= b <0 vs -1.19209e-007>

在谷歌了很多方法后,才解决了问题,目前的最好办法只能规避这个问题,如下:

1.找到math_functions.cpp,并打开路径为~/caffe/src/caffe/util的math_functions.cpp

2.找到第250行出错的地方,双斜杠屏蔽      // CHECK_LE(a, b);

3.重新在caffe路径下make py




如果注释掉 CHECK_LE(a, b) 会出现Data layer prefetch queue empty

不注释CHECK_LE(a, b) 会出现错误 a可能大于b

这种问题出现通常是注释掉 CHECK_LE(a, b) 出现Data layer prefetch queue empty。导致程序出现死循环。

解决办法修改src/caffe/util/sampler.cpp,如下面修改代码所示//renew注释下,加入两个判断,使得bbox长宽不要越界。

  1. // Figure out bbox dimension.
  2.   float bbox_width = scale * sqrt(aspect_ratio);
  3.   float bbox_height = scale / sqrt(aspect_ratio);



  4.   //renew
  5.   if(bbox_width>=1.0)
  6.   {
  7.     bbox_width=1.0;
  8.   }
  9.   if(bbox_height>=1.0)
  10.   {
  11.     bbox_height=1.0;
  12.   }


  13.   // Figure out top left coordinates.
  14.   float w_off, h_off;
  15.   caffe_rng_uniform(1, 0.f, 1 - bbox_width, &w_off);
  16.   caffe_rng_uniform(1, 0.f, 1 - bbox_height, &h_off);
复制代码

注意:修改cpp文件后需要重新在caffe路径下make py


问题:Caffe-SSD训练过程中出现blocking_queue.cpp:50] Data layer prefetch queue empty

原因:SSD训练慢的原因是 在生成LMDB文件时没有将图片转化为300*300分辨率大小 所以在训练网络时需要转换图片分辨率 这样就造成了数据输入层预存队列为空
GPU一直在等待数据中 所以造成了训练异常缓慢。
(图片resize是在CPU上运行的,CPU数据读入慢于GPU计算,导致GPU处于空闲,使得计算速度下降,迭代时间间隔扩大。)

解决办法:找到生成LMDB格式数据的文件create_data.sh ,将width=0改为width=300,将height=0改为height=300.

注意:直接读入图片的是使用ImageDataLayer ,使用LMDB格式的数据是用DataLayer。


内存不够的错误:Check failed: error == cudaSuccess (2 vs. 0)  out of memory



修改:ssd_face_dfy.py  中 batch_size = 8    与   accum_batch_size =8   尽量小点  原来是32







make py.png
让天下人人学会人工智能!人工智能的前景一片大好!
回复

使用道具 举报

1365

主题

1856

帖子

1万

积分

管理员

Rank: 10Rank: 10Rank: 10

积分
14418
QQ
 楼主| 发表于 2019-8-6 09:15:15 | 显示全部楼层
之后在虚拟环境终端中运行 python examples/ssd/ssd_face_dfy.py
报错:math_functions.cpp:250] Check failed: a <= b (0 vs. -1.19209e-07)
*** Check failure stack trace: ***
    @     0x7f313f9100cd  google:ogMessage::Fail()
    @     0x7f313f911f33  google:ogMessage::SendToLog()
    @     0x7f313f90fc28  google:ogMessage::Flush()
    @     0x7f313f912999  google:ogMessageFatal::~LogMessageFatal()
    @     0x7f314001dfe7  caffe::caffe_rng_uniform<>()
    @     0x7f313ffc1358  caffe::SampleBBox()
    @     0x7f313ffc16b0  caffe::GenerateSamples()
    @     0x7f313ffc1900  caffe::GenerateBatchSamples()
    @     0x7f314013b602  caffe::AnnotatedDataLayer<>::load_batch()
    @     0x7f314006059a  caffe::BasePrefetchingDataLayer<>::InternalThreadEntry()
    @     0x7f3140032b15  caffe::InternalThread::entry()
    @     0x7f313350abcd  (unknown)
    @     0x7f311ddab6db  start_thread
    @     0x7f313e05388f  clone
Aborted (core dumped)
让天下人人学会人工智能!人工智能的前景一片大好!
回复

使用道具 举报

0

主题

117

帖子

258

积分

中级会员

Rank: 3Rank: 3

积分
258
QQ
发表于 2020-2-3 15:48:34 | 显示全部楼层
谢谢老师提供的资料。
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则

QQ|Archiver|手机版|小黑屋|人工智能工程师的摇篮 ( 湘ICP备2020019608号-1 )

GMT+8, 2024-3-29 21:58 , Processed in 0.197147 second(s), 22 queries .

Powered by Discuz! X3.4

© 2001-2017 Comsenz Inc.

快速回复 返回顶部 返回列表