GluonCV 0.2


#1

blog link: https://zh.mxnet.io/blog/gluon-cv-0.2


#2

沐神这是不给OpenCV活路啊:rofl:


#3

请问能不能在文档中提供预训练模型在各数据集上的精度数值呢?


#4

请问,gluoncv 0.2 中复现的PSPNet在ADE20K上达到了多少精度?我看一些参数的设置并不是太合理,感觉不是很正确的复现呀。


#5

支持!顺便把yolov3做了吧~ :grinning:


#6

换了6/26上传的train脚本,在练Faster-RCNN时出现了:
mxnet.base.MXNetError: Cannot find argument ‘static_alloc’, Possible Arguments: 这个错误
环境:Mxnet1.2 GlounCV 0.2 (6/26 update)


#7

错误信息能贴全嘛…


#8

Traceback (most recent call last):
File “D:/mxnet-test/train2.py”, line 422, in
train(args)
File “D:/mxnet-test/train2.py”, line 364, in train
cls_pred, box_pred, roi, samples, matches, rpn_score, rpn_box, anchors = net(data, gt_box)
File “C:\Anaconda3\lib\site-packages\mxnet\gluon\block.py”, line 413, in call
return self.forward(*args)
File “C:\Anaconda3\lib\site-packages\mxnet\gluon\block.py”, line 621, in forward
return self._call_cached_op(x, *args)
File “C:\Anaconda3\lib\site-packages\mxnet\gluon\block.py”, line 522, in _call_cached_op
self._build_cache(*args)
File “C:\Anaconda3\lib\site-packages\mxnet\gluon\block.py”, line 483, in _build_cache
self._cached_op = ndarray.CachedOp(out, self._flags)
File “C:\Anaconda3\lib\site-packages\mxnet_ctypes\ndarray.py”, line 115, in init
ctypes.byref(self.handle)))
File “C:\Anaconda3\lib\site-packages\mxnet\base.py”, line 149, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: Cannot find argument ‘static_alloc’, Possible Arguments:

inline_limit : int (non-negative), optional, default=2
Maximum number of operators that can be inlined.
forward_bulk_size : int (non-negative), optional, default=15
Segment size of bulk execution during forward pass.
backward_bulk_size : int (non-negative), optional, default=15
Segment size of bulk execution during backward pass.

Process finished with exit code 1


#9
  1. Mxnet版本号多少
  2. 尝试pull gluon-cv,最近有更新
  3. 给出复现代码

#10

期待cityscape 训练集的加入


#11

static alloc可能需要更新的mxnet版本。


#12

https://gluon-cv.mxnet.io/model_zoo/index.html#image-classification


#13

mIOU 41.1. 比原文差0.5 应该是base network 造成的


#14

哎呀,之前没注意到文档这一块,谢谢


#15

请问一下,resnet50_v1b 和 resnet50_v2a 相较于原版的resnet,做了那些改动?


#16

支持支持,呼唤yolo3:smiley:。顺便问问resnet50_V1b改动是BN的gamma初始化么?


#17

unresolved reference ‘gluoncv’ , 这个怎么破?


#18

解决了, 抱歉


#19

想请教一下label分配id的问题,在0.2 版本里面,voc数据集id 编号是从0-20 还是1-21啊


#20

@zhreshold