ncnn
ncnn copied to clipboard
ncnn is a high-performance neural network inference framework optimized for the mobile platform
MAC本版本:MacBook Pro (Retina, 13-inch, Early 2015) 系统版本12.5 NCNN版本:ncnn-20220729-full-source 关闭openmp编译选项 执行 ``` mkdir -p build-ios cd build-ios cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/ios.toolchain.cmake -DIOS_PLATFORM=OS -DIOS_ARCH="armv7;armv7s;arm64;i386;x86_64" .. make -j 4 ``` 提示以下错误 ``` ...... [ 42%]...
## detail | 详细描述 | 詳細な説明 在安卓端跑模型推理时不禁用use_fp16_arithmetic情况下,第一层卷积层部分输出就为0,pc端则不会,请问各位有遇到类似问题吗,求解?? 设备:Qualcomm Snapdragon 778G Plus 模型:模型均为采用ncnnoptimize优化的fp16模型 (onnx2ncnn导出模型问题一致) 安卓端在不禁用use_fp16_arithmetic情况下第一层卷积层部分输出: (左边是安卓端,右边是pc端)  禁用use_fp16_arithmetic情况下,精度相差极少: 
error: ``` foldable_constant output_mean.1 libc++abi: terminating with uncaught exception of type c10::Error: Tensors of type TensorImpl do not have sizes Exception raised from sizes_custom at /Users/runner/work/pytorch/pytorch/pytorch/c10/core/TensorImpl.cpp:416 (most recent call first):...
## detail | 详细描述 | 詳細な説明 Basically a conclusion from here: https://github.com/TabuaTambalam/DalleWebms/issues/1 I made this cuz I can't find any doc about using NCNN with vulkan in python, and I'm...
#### 算法思路 堆排序`partial_sort`算法思路如下: 一个大小为n的array,我们要获得`top k`个最大(最小)的元素。 * 以array的前k个元素建立一个大小为k的小(大)根堆(使用自定义`heapify()`函数) * 遍历剩余`n-k`个元素与小(大)根堆的堆顶元素比较,如果比堆顶元素大(小)那么就会交换两者同时重新更新小(大)根堆,遍历结束后会获得`top k`个最大(最小)的元素,但是并不是按照严格的顺序来排序。 * 利用堆顶是最大(最小)的元素对这k个元素使用常规意义下的堆排序来依次获得严格降序(升序)的`top k` array. *以上所有操作均为inplace操作* #### 复杂度分析 * 时间复杂度 之前冒泡排序时间复杂度为`O(nk)`, 此处堆排序时间复杂度为`O((n-k)logk)`。 * 空间复杂度 空间复杂度相同,均为`inplace`操作并没有使用额外空间。
## error log | 日志或报错信息 | ログ terminate called after throwing an instance of 'c10::Error' what(): forward() is missing value for argument 'r1'. Declaration: forward(__torch__.model.model.MattingNetwork self, Tensor src, Tensor r1,...
## error log | 日志或报错信息 | ログ [ 99%] Building CXX object python/CMakeFiles/pyncnn.dir/src/main.cpp.o In file included from /data/data/com.termux/files/home/ncnn/python/src/main.cpp:16: In file included from /data/data/com.termux/files/home/ncnn/python/pybind11/include/pybind11/pybind11.h:13: In file included from /data/data/com.termux/files/home/ncnn/python/pybind11/include/pybind11/attr.h:13: In file...
* 实现了 int8 layernorm,imagenet1k 只量化 LN 精度 84.01->83.68 *
## error log | 日志或报错信息 | ログ [Layer wise time taken during inference.](https://bin.wantguns.dev/p/xDJvxx.txt) ## context | 编译/运行环境 | バックグラウンド By compiling the code with benchmark flag, we can observe that...
## error log | 日志或报错信息 | ログ Segmentation fault on `ncnn::Extractor::extract` when using multi-input model. NCNN version: [ncnn-20220729-ubuntu-2004.zip](https://github.com/Tencent/ncnn/releases/download/20220729/ncnn-20220729-ubuntu-2004.zip) ## context | 编译/运行环境 | バックグラウンド ## how to reproduce | 复现步骤...