small bird

关于使用mnn量化模型之后，完全检测不到目标的疑问

1

大佬，我最近在用nanodet工程里面的demo_mnn代码。 临时使用的是demo_mnn里面的python脚本做测试的。发现了一个问题：mnn将Onnx转成.mnn模型后，可以直接检测到结果的，符合预期。但是，使用mnn进行int8量化之后，就完全检测不到结果了。 我确定自己的转换过程是正确的，我特意试了一下yolov5s模型，量化后是正常检测到目标的。 不清楚群主是否试验过。我用的backbone是shufflenetV2。

unable to init server: Could not connect: Connection refused

3

我在训练过程中，总是出现“unable to init server: Could not connect: Connection refused”这个报错。详细报错见下面。 /usr/local/lib/python3.6/dist-packages/pytorch_lightning/metrics/__init__.py:44: LightningDeprecationWarning: `pytorch_lightning.metrics.*` module has been renamed to `torchmetrics.*` and split off to its own package (https://github.com/PyTorchLightning/metrics) since v1.3 and will...

IndexError: The shape of the mask [2101120] at index 0 does not match the shape of the indexed tensor [656600] at index 0

2

When I run reppoints_v2_r50_fpn_1x_coco.py and reppoints_v2_r50_fpn_giou_1x_coco.py, the program can run normally, no error is reported, and the loss value is as expected. However, when I run reppoints_v2_r101_fpn_dconv_c3-c5_giou_mstrain_2x_coco.py, reppoints_v2_r101_fpn_giou_mstrain_2x_coco.py, reppoints_v2_x101_fpn_dconv_c3-c5_copn_dconv_c3-c5_giou_train_giou_train_mstrain_2x_coco.py, and...

使用cmake .. 命令之后提示有错误。

1

visionlab@visionlab:~/workspace/liuqinglong/ncnn-mobilenet-ssd-master/ncnn-mobilenet-ssd/build$ cmake .. -- CMAKE_INSTALL_PREFIX = /home/visionlab/workspace/liuqinglong/ncnn-mobilenet-ssd-master/ncnn-mobilenet-ssd/build/install CMake Error at /usr/local/share/OpenCV/OpenCVConfig.cmake:205 (message): opencv_imgcodecs is required but was not found Call Stack (most recent call first): ssdmobilenet/CMakeLists.txt:2 (find_package) -- Configuring incomplete,...