小狗hammer

小米安卓demo，MNN::CV::ImageProcess::convert报错崩溃无法运行

4

# 平台(如果交叉编译请再附上交叉编译目标平台): # Platform(Include target platform as well if cross-compiling): 安卓，小米黑鲨4 # Github版本: # Github Version: MNN 1.2.0-release， 从[这里](https://github.com/alibaba/MNN/releases)下载的 commit-id : 58545d6ca142682424dfd1eed578a8393b81b76e # 编译方式: # Compiling Method ``` mkdir build_64...

测试自己的模型安卓的benchmark卡住

1

**1. 环境（environment）** - Build OS and Version: Ubuntu - RunTime OS Version: Android - RunTime DEVICE: ARM/OPENCL **2. Github版本** - branch：master - commit(optional): 9766eeedb2734cb5231 **3. 编译方式(compile method)** cd benchmark/benchmark_android ./benchmark_models.sh...

Release tag=1.20.0 has version conflict

2

When I run this command `python3 -m pip install AimetTensorflow-tf_cpu_1.20.0-cp36-cp36m-linux_x86_64.whl` I get the following error The conflict is caused by: aimettensorflow tf-cpu-1.20.0 depends on h5py==2.9.0 tensorflow 2.4.0 depends on h5py~=2.10.0...

The results of QuantizationSimModel and DSP deployment are different

3

**Definition of my model** `class Net(nn.Module): def __init__(self): super(Net, self).__init__() self.avg_pool = nn.AdaptiveAvgPool2d(1) self.fc1 = nn.Linear(3, 256) self.fc2 = nn.Linear(256, 10) def forward(self, x): # (1,3,20,20) x = self.avg_pool(x) x...

I download BLseresnext50-32x4d weights file,and it mention that Compressed package is corrupted. For windows, i try 7-zip tools to compress, its failed.And Then i try it in ubuntu16.04, it also...

issues about Mix-Precision training

1

Thanks for your great work!! Because of the limitation of GPU, I used pytorch native amp for training. However, I found some problems with numerical stability while computing giou, NaN...

您好！ 我的硬件平台是3588 android，运行代码过程中我连接了RK开发板 我在尝试导出rk_batch_size=2的rknn模型，并用此模型执行accuracy_analysis 在执行accuracy_analysis的时候遇到错误 **下面几种情况都会报错，如果rknn_batch_size=1，那么一切都会正常** 这是RK导出代码 `ret = rknn.build(do_quantization=True, dataset=str(OUTPUT_DIR / "datasets.txt"), rknn_batch_size=2)` 这是analysis代码，我尝试了多种 `ret = rknn.accuracy_analysis(inputs=["1.png",], output_dir='./snapshot', target='rk3588')` `ret = rknn.accuracy_analysis(inputs=["1.png","2.png"], output_dir='./snapshot', target='rk3588')` 我同样把两张图像做成了shape为[2,h,c,3]的格式保存为npy，并执行 `ret = rknn.accuracy_analysis(inputs=["1.npy",], output_dir='./snapshot',...

大佬，您好！ 感谢这个项目的贡献，对我这种刚入门的非常友好！ 对于raycast里面的get_min_time和get_max_time，两个函数感觉非常困惑。 初始化： ``` // 第一帧的相机位姿设置在 Volume 的中心, 然后在z轴上拉远一点 current_pose.setIdentity(); current_pose(0, 3) = _configuration.volume_size.x / 2 * _configuration.voxel_scale; current_pose(1, 3) = _configuration.volume_size.y / 2 * _configuration.voxel_scale; current_pose(2, 3) =...