search

zhreshold / mxnet-ssd

MXNet port of SSD: Single Shot MultiBox Object Detector. Reimplementation of https://github.com/weiliu89/caffe/tree/ssd
MIT License
764 stars 337 forks source link

AttributeError: 'str' object has no attribute 'rank' #224

Open TonyTangYu opened 5 years ago

TonyTangYu commented 5 years ago

I run python train.py --gpu 0 --batch-size 32 --lr 0.001 but I get the following problem.

Using mxnet as: <module 'mxnet' from '/THL5/home/daodao/softwares/mxnet0.11/python/mxnet/init.pyc'> Warning: using pre-installed version of mxnet may cause unexpected error... (export MXNET_EXAMPLE_SSD_DISABLE_PRE_INSTALLED=1) to prevent loading pre-installed mxnet. [22:12:49] src/io/iter_image_det_recordio.cc:280: ImageDetRecordIOParser: /THL5/home/daodao/ty_project/model/mxnet-ssd/data/train.rec, use 27 threads for decoding.. [22:12:50] src/io/iter_image_det_recordio.cc:333: ImageDetRecordIOParser: /THL5/home/daodao/ty_project/model/mxnet-ssd/data/train.rec, label padding width: 350 [22:12:50] src/engine/threaded_engine_perdevice.cc:253: [added by cxt] engine is created by EnginePerDevice [22:12:50] src/io/iter_image_det_recordio.cc:280: ImageDetRecordIOParser: /THL5/home/daodao/ty_project/model/mxnet-ssd/data/val.rec, use 27 threads for decoding.. [22:12:51] src/io/iter_image_det_recordio.cc:333: ImageDetRecordIOParser: /THL5/home/daodao/ty_project/model/mxnet-ssd/data/val.rec, label padding width: 350 INFO:root:Start training with (gpu(0)) from pretrained model /THL5/home/daodao/ty_project/model/mxnet-ssd/model/vgg16_reduced [22:12:52] src/nnvm/legacy_json_util.cc:190: Loading symbol saved by previous version v0.8.0. Attempting to upgrade... [22:12:52] src/nnvm/legacy_json_util.cc:198: Symbol successfully upgraded! INFO:root:Freezed parameters: [conv1_1_weight,conv1_1_bias,conv1_2_weight,conv1_2_bias,conv2_1_weight,conv2_1_bias,conv2_2_weight,conv2_2_bias] [22:13:03] src/operator/././cudnn_algoreg-inl.h:112: Running performance tests to find the best convolution algorithm, this can take a while... (setting env variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable) [22:13:40] src/operator/convolution.cu:119: This convolution is not supported by cudnn, MXNET convolution is applied. ff Traceback (most recent call last): File "train.py", line 148, in tensorboard=args.tensorboard) File "/THL5/home/daodao/ty_project/model/mxnet-ssd/train/train_net.py", line 354, in train_net monitor=monitor) File "/THL5/home/daodao/softwares/mxnet0.11/python/mxnet/module/base_module.py", line 568, in fit rank=kvstore.rank, total_iter=temp_count, AttributeError: 'str' object has no attribute 'rank'

It seems that it results from the mxnet. There is no solution corresponding to this problem. I wanna solve this problem but don't know how. Could you please help me get over with it? Thank you!

zhreshold commented 5 years ago

please check mxnet version, make sure it's 1.1 or 1.2

TonyTangYu commented 5 years ago

please check mxnet version, make sure it's 1.1 or 1.2

Thank you for your response. I checked the version of mxnet. It's 0.11. What can I do to solve this problem?

TonyTangYu commented 5 years ago

@zhreshold In fact, I downloaded the source code of mxnet 0.11 and try to train vgg16_reduced but got the same error. Is it a problem from the base_module.py or resulting from the change of mxnet's API?

zhreshold commented 5 years ago

@TonyTangYu I mean mxnet version is 1.1 not 0.11

TonyTangYu commented 5 years ago

@zhreshold I know. Do you mean I should install mxnet 1.1 or 1.2? However I want to know how to solve this problem when it's 0.11. Could you do me a favour?

zhreshold commented 5 years ago

@TonyTangYu 

pip install mxnet==1.2.0 --upgrade
TonyTangYu commented 5 years ago

@zhreshold Thank you for your response. But I install mxnet0.11 on the cloud server. It cannot connect the Internet and the pip command won't work on this occasion. Could you please give me another solution? Thank you.

TonyTangYu commented 5 years ago

@zhreshold I upgraded the version of MXNet. It works now. Thank you!