xiaolai-sqlai / gluon-reid Goto Github PK

thx for your job!
this is my cmd:
python train.py --num-gpus 1 --batch-size 32 --epochs 20,40,60
the other setings are default.
this is my eval results:
top1:0.367041 top5:0.561798 top10:0.655431 mAP:0.269477
why?
pls help me!

my mistake

The program broke when it goes to "trainer.step(batch_size)",and the console prints:
mxnet.gluon.parameter.DeferredInitializationError: Parameter 'resnet0_dense0_weight' has not been initialized yet because initialization was deferred. Actual initialization happens during the first forward pass. Please pass one batch of data through the network before accessing Parameters. You can also avoid deferred initialization by specifying in_units, num_features, etc., for network layers.

How to fix it,please?

https://github.com/xiaolai-sqlai/gluon-reid/blob/master/branch/data_read/sampler.py#L35
这里的len是不是应该返回self.num_samples？也就是_iter_函数中ret的长度

我看https://mxnet.incubator.apache.org/_modules/mxnet/gluon/data/sampler.html#BatchSampler
batchsampler的源码返回的是batch的数目

when opt.hybridize=True
AssertionError: Argument data must have NDArray type, but got Symbol resnet0_hybridsequential2_activation7
in the line: parts_2 = nd.split(x2, axis=2, num_outputs=2)

Cool project! Just to fix typo in proj descriptions

many STOA algorithm.
->
many SOTA algorithms.

hi，thanks for sharing，but I do not get the results with the code in branch folder，it get an error at the extract_feature function in test code。

I have searched through the code and am unable to find your code for the MGN network although you report results. Does your code not exist in the repo yet and if so when might it?

accuracy in baseline module is not the accuracy of all the validation set, but the accuracy of the last batch of validation set.