lance
lance
## 测试过程 ### 1. 国产芯片测试 1. 使用mmcls 训练mobilenetv2_1.5模型并转为onnx 2. 下载[autoslim](https://download.openmmlab.com/mmrazor/v0.1/pruning/autoslim/autoslim_mbv2_subnet_8xb256_in1k/autoslim_mbv2_subnet_8xb256_in1k_flops-0.22M_acc-71.39_20211222-43117c7b.pth)，使用mmdeploy自带的torch2onnx转为onnx，其中相关mcfg如下： ```python # deploy_cfg onnx_config = dict( type="onnx", export_params=True, keep_initializers_as_inputs=False, opset_version=10, save_file="mobilenetv2_mmdeploy", input_names=["input"], output_names=["output"], input_shape=[224, 224], ) backend_config = dict(type="onnxruntime") #...
Thank you for the guidance, I will look at the source code recently and try to fix it
@HIT-cwh https://github.com/open-mmlab/mmrazor/pull/193 - origin mobilenet_v2_1.5 39.25MB - search config ```python _base_ = [ "./supernet.py", ] algorithm = dict(distiller=None, input_shape=(3, 224, 224)) searcher = dict( type="GreedySearcher", target_flops=[300000000, 200000000], max_channel_bins=12, metrics="accuracy", metric_options={"topk":...
oh no, export onnx weight size are equal
get it!tks
@wolfworld6 你好，我和你遇到了同样的问题，请问您解决了嘛
@fqlovetb @ai815 请问你们转换后的效果怎么样
Have you detected the target of the training custom dataset? What is the format of your data, my loss all Nan
@Scalsol TKS
@Scalsol TKS help me _base_ = './reppoints_v2_r50_fpn_giou_1x_coco.py' # learning policy lr_config = dict(step=[16, 22]) total_epochs = 24 # multi-scale training img_norm_cfg = dict( mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)...