kingfengji / deepmiml Goto Github PK

Hi,
Greateful github contribution!
But I have a concern about the variable 'n_isntances' in deep_MIML.py.
According to the paper, shouldn't the n_isntances be the depth of the 3D tensor?
I found you use H*W be the 'n_isntances'.

I tried this projects on ubuntu16.04 , but I had this question: raise RuntimeError('To use MKL 2018 with Theano you must set "MKL_THREADING_LAYER=GNU" in your environement')

你好，提供的vgg权重文件打不开，麻烦可以重新提交一下吗。

Softmax is applied on the labels axis, but not specify in paper.
Relu is not applied on conv(1,1), but said used in paper.

Thus the paper seems not matched and not reproducible, can you explain?

I want to use my own datasets to train this model, can anyone provide me some tutorials about the datasets?

I want to train a model on the multi-instance multi-label dataset (scene, text), but it does not optimize. So i wonder if there is a demo that i can borrow to bulid a model on scene and text datasets

why adding 8 in the formula?
$(x, y)=\left(\operatorname{con} v_{x} * 16+8, \operatorname{con} v_{y} * 16+8\right)$

I have ran the code, and it's very slow, the ETA is more than 600. I want to know others need how long to run the code.
I wonder how to change the code to use the gpu. Thank you.

I thought it should be # shape -> (n_bags, (L * K), H, W)
Right?