Oct. 12, 2019, 3:33 p.m.
read: 40
准备linux 以UBUNTU16.01为例 下载git clone https://github.com/kaldi-asr/kaldi.git or wget https://github.com/kaldi-asr/kaldi/archive/master.zip unzip master.zip mv kaldi-master kaldi 检查依赖cd kaldi/tools ./extras/check_dependencies.sh 直到输出 ./extras/check_dependencies.sh: all OK. 编译tools目录确认 当前工作目录还...
Oct. 3, 2019, 3:51 p.m.
read: 34
传入参数为已经实例化完成的网络，输出为所有需要求梯度的参数的数量 def count_parameters(model): return sum(p.numel() for p in model.parameters() if p.requires_grad)
Sept. 23, 2019, 5:51 p.m.
read: 32
导出wav ffmpeg -i huoying.mp4 -f wav -ar 16000 huoying.wav
Sept. 19, 2019, 3:36 p.m.
read: 20
论文的题目是: A Time Delay Neural Network with Shared Weight Self-Attention forSmall-Footprint Keyword Spotting 来自中科院一篇文章https://www.isca-speech.org/archive/Interspeech_2019/pdfs/1676.pdf 作者为白烨,分享一篇他的综述https://mp.weixin.qq.com/s/c-lQETCPQTztZjV4TYTsZA 这个文章中提出了一个新的网络，整个网络的参数量仅12K,适合在移动端进行相关部署。 这个网络中未使...
Sept. 2, 2019, 6:16 p.m.
read: 79
ASSERT FAILED at /pytorch/aten/src/ATen/native/cudnn/RNN.cpp:528 RuntimeError: params_from.size(0) == params_to.size(0) ASSERT FAILED at /pytorch/aten/src/ATen/native/cudnn/RNN.cpp:528, please report a bug to PyTorch. number of layers mismatch 这个bug仅在1.0.0上出现,在它的上一个版本 0.4.1与下一个版本1.1.0中都未复现,所以解决方...
Aug. 29, 2019, 3:30 p.m.
read: 62
这里有两个重要的前提:1.当list长度为1,仅仅是一个元素的时候,认为是相等2.list长度不为0 def check_same(l): if len(set(l)) == 1: return True else: return False a = [1, 1, 1, 1] b = [1, 2, 3, 4] print(check_same(a)) print(check_same(b)) 输出结果为: TrueFalse
Aug. 21, 2019, 5:47 p.m.
read: 65
发生 错误 UnicodeDecodeError: ‘utf8’ codec can’t decode byte 0xc0 in position 3: invalid start byte 原因:sppas目录不能含有中文
Aug. 18, 2019, 12:25 p.m.
read: 88
数据来自ssh蜜罐收集:共23639条,下面是部分数据 root/qyeee1234 root/cnnic root/utterly root/broke root/963256478715 root/whereon root/quick root/cdy123 root/iammecn@weiphone root/19800730 root/19740712 root/prince root/abdcefghi root/roping root/19790519 root/19730622 root/passroot root/wxmsyy42410...
Aug. 9, 2019, 10:27 p.m.
read: 87
0x00 交叉熵Loss(Softmax Loss) import torch import torch.nn as nn import torch.nn.functional as F # 自己实现一个LossSoftmaxLoss #可以看到 class LogSoftmaxLoss: def _onehot_encoding_(self, label, num_classes): shape_0 = label.shape[0] label = label.reshape(-1, 1) result = torch.zero...
July 12, 2019, 11:39 a.m.
read: 115
浏览器UA标识收集计划 以下UA从某网站的真实访问数据筛选去重而来包含操作系统信息的UA(1163条) Mozilla/5.0 (iPhone; CPU iPhone OS 11_0 like Mac OS X) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/47.0.2648.72 Mobile Safari/537.36 Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/73.0.3683.103 Safari/5...
June 19, 2019, 6:26 p.m.
read: 80
>> subplot(231);imagesc(model_0) >> subplot(232);imagesc(model_1) >> subplot(233);imagesc(model_2) >> subplot(234);imagesc(model_3) >> subplot(235);imagesc(model_4) >> subplot(236);imagesc(model_5)
June 19, 2019, 3:54 p.m.
read: 64
June 18, 2019, 5:33 p.m.
read: 119
转自https://discuss.pytorch.org/t/how-to-reverse-a-torch-tensor/382 形状为（10）的tensor翻转 tensor = torch.rand(10) # your tensor # create inverted indices idx = [i for i in range(tensor.size(0)-1, -1, -1)] idx = torch.LongTensor(idx) inverted_tensor = tensor.index_select(0, idx) 形状为（1，10，1）的tensor翻转 tens...
June 17, 2019, 7:36 p.m.
read: 75
以下demo只支持float格式，仅提供思路，其他数据类型需自行扩展： import numpy as np import struct def bin2npy(bin_file_path, npy_file_path, type='float'): assert type in ['float'] out_data = [] with open(bin_file_path, mode='rb') as bin_file: in_data = bin_file.read() for i in range(len(in_data)//...
June 12, 2019, 5:33 p.m.
read: 67
1. 添加权限到 AndroidManifest.xml2. 添加以下语句到build.gradle=>dependenciesimplementation 'com.mylhyl:acp:1.2.0' 然后sync project 3.添加Java语句以动态请求权限Acp.getInstance(this).request(new AcpOptions.Builder() .setPermissions( Manifest.permission.RECORD_AUDIO...
June 10, 2019, 9:10 p.m.
read: 109
适合最后一层（输出层）为二分类,label==>1 为一个类别，label==> 0 为另一个类别两个样例： >>> target = torch.ones([10, 64], dtype=torch.float32) # 64 classes, batch size = 10 >>> output = torch.full([10, 64], 0.999) # A prediction (logit) >>> pos_weight = torch.ones([64]) # All weights are equal...
Feb. 22, 2019, 4 p.m.
read: 62
Android使用VoIP音频链路 AudioManager audioManager = (AudioManager) getSystemService(Context.AUDIO_SERVICE); audioManager.setMode(AudioManager.MODE_IN_COMMUNICATION); audioManager.setSpeakerphoneOn(true);//true 底部喇叭播放，false 顶部喇叭
Feb. 3, 2019, 1:02 a.m.
read: 123
监控在安保和巡查中发挥着重要作用，但也是一项非常乏味的任务，深度学习的出现在一定程度上将人类从这一任务中解放出来。本项目基于深度学习的目标检测去搭建了一个简单有效的监控系统，能够自动化进行人流统计和行人检测。项目地址：https://github.com/zhangpengpengpeng/PedestrianDetectionSystem