search

gujiuxiang / MIL.pytorch

PyTorch implementation of Multiple-instance learning
117 stars 29 forks source link

No such file model/coco_valid1_eval.pkl #7

Open SilencerChen opened 6 years ago

SilencerChen commented 6 years ago

I try to run the test code but couldn't find model/coco_valid1_eval.pkl file. Could you updata it ? Thanks so much.

YuanEZhou commented 6 years ago

I have the same issue. @SilencerChen @gujiuxiang

gujiuxiang commented 6 years ago

@SilencerChen @YuanEZhou, uploaded, pls find the file in Weiyun. Thanks.

winnechan commented 5 years ago

i have create a pkl file that can be loaded without coco_voc by only storing the .details.precision details.score numpy.array.

https://drive.google.com/open?id=1eRrH45xOj41__v2bXMVa8OGZ1Gj6NYze

using python 3 to load the python 2 pkl file:

with open(pickle_file_name, 'rb') as f:   d = cPickle.load(f, encoding='latin1')

but there's a problem i don't know why: i can get reasonable results by removing the softmax layer in the model.

another problem is that, to me, the output before softmax seems to be the noisy-OR pooling of each image region instance, which can be used to be the final output directly without the score matching of unknown meaning.

can someone explain?