Working on Windows with same accuracy

[Ram-Godavarthi]

Hi, Is it possible to run the code on windows.? Is it similar to the original faster rcnn in terms of accuracy ? Did you do inference with this repo??

[WalterMa]

Yes, I writed and tested on windows. But currently, the mAP on voc dataset is lower than original faster rcnn... 😭

[Ram-Godavarthi]

Okay Fine. Is the procedure holds same for both windows and ubuntu??

[Ram-Godavarthi]

I am getting this error when running demo script in windows..

File "demo_faster_rcnn.py", line 59, in bbox_mean=cfg.bbox_mean, bbox_std=cfg.bbox_std) File "C:\Users\godavart\Desktop\gluon-faster-rcnn-master\rcnn\rcnn.py", line 37, in init negative_iou_threshold=rpn_negative_threshold) File "C:\Users\godavart\Desktop\gluon-faster-rcnn-master\rcnn\anchor.py", line 13, in init self._base_anchors = self.params.get_constant('base_anchors', base_anchors) AttributeError: 'ParameterDict' object has no attribute 'get_constant'

[WalterMa]

Could you provide your mxnet version? Seems it is caused of your mxnet is out of date.

[Ram-Godavarthi]

Even i thought the same. In windows it is 1.0.0.. How to upgrade this in windows using command line? any command for that

[WalterMa]

pip install --upgrade --pre mxnet-cuXX

XX is your cuda version, eg. mxnet-cu80 if you use cuda 8.0

Refer to https://pypi.org/project/mxnet/ for all aviliable packages.

[Ram-Godavarthi]

I could able to run it on aws. I will try on windows later.. I want it to work fine in aws now.. Later i will play with it in windows.

[Ram-Godavarthi]

Which Pretrained model should i use to check the demo.? Can You provide me the link?

I have used the one from here I used vgg16 param file . But i am getting an error something like this .

raceback (most recent call last): File "./demo_faster_rcnn.py", line 61, in net.load_params(cfg.model_params.strip(), ctx=ctx) File "/home/ubuntu/anaconda3/lib/python3.6/site-packages/mxnet/gluon/block.py", line 340, in load_params filename, ctx, allow_missing, ignore_extra, self.prefix) File "/home/ubuntu/anaconda3/lib/python3.6/site-packages/mxnet/gluon/parameter.py", line 769, in load "with '%s'"%(restore_prefix, name, restore_prefix) AssertionError: restoreprefix is 'fasterrcnn0' but Parameters name 'vgg0_conv0weight' does not start with 'fasterrcnn0'

Help me out in this..

[WalterMa]

Pertained model parameters are not release d yet, since the mAP on VOC is lower than original at this config.

[WalterMa]

The model file you referred is the base model used to init vgg feature extract layers that will be downloaded automatically in the frist you trian.

[Ram-Godavarthi]

I got it.. Thank You.. I am not using any pretrained model, instead i am randomizing the weights...