search

goddoe / hide-and-seek

Tensorflow implementation of "Hide-and-Seek: Forcing a Network to be Meticulous for Weakly-supervised Object and Action Localization"[ICCV 2017]
12 stars 4 forks source link

download_data.py and reference issues #1

Open sarcinelliRenan opened 6 years ago

sarcinelliRenan commented 6 years ago

The download_data.py does not seems to be downloading the data. I tried to download the dataset from the internet but it doesn't contains the meta.pickle file. The paths from the scripts are also inconsistent, the download_data.py saves the files on a tmp folder, but the other scripts are referring tho the data folder. The load_and_test_HaS.ipynb tries to import the load_data_with_meta and load_meta functions from helpers but this functions are not implemented.

goddoe commented 6 years ago

Hello @sarcinelliRenan . Thanks for reporting. I'll check and fix the issue soon. Thanks! :)

goddoe commented 6 years ago

Hi again, @sarcinelliRenan . I fixed the issue that you reported. However, The codes are too old for now and there are many things that needed to be refactored or update. Therefore, I'll continue update my codes. thanks!

sarcinelliRenan commented 5 years ago

Hello @goddoe . I noticed that there are some "bugs", probably outdated code, and I was trying to fix it myself. I'm tying to use the Alexnet_v2 model, but I didn't found a checkpoint file for this network, then I started training it, but the loss is stocked in 5, and the accuracy is almost the same as random. Do you have any pre-trained model to provide. Anyway, thanks for the attention and the fixes.

goddoe commented 5 years ago

@sarcinelliRenan Sorry, there is no pretrained model for alexnet_v2. There are pretrained model for InceptionV3 and I added some codes for downloading InceptionV3 at download_data.py. Try 'InceptionV3' as parameter 'model_base_name', when you make Detector instance.