xiaoyuan1996 / amfmn Goto Github PK

Hello, I have two questions I want to ask you, what is the training resource required for it (graphics card type, duration)，I have a Titan XP (12G), how long will it take to train with it?

MIssing -- seq2vec # some files about seq2vec
-- bi_skip.npz
-- bi_skip.npz.pkl
-- btable.npy
-- dictionary.txt
-- uni_skip.npz
-- uni_skip.npz.pkl
-- utable.npy

Hello,I am very interested in your paper "Exploring a Fine-Grained Multiscale Method for Cross-Modal Remote Sensing Image Retrieval". In your paper, you designed a triple loss function with dynamic variable margin, but I can't find this relevant code in the code. Can you tell me where it is?

Hello,
I'm trying to execute your code but I'm getting errors:

Current lr: 0.0002
Traceback (most recent call last):
File "train.py", line 209, in
main(update_options)
File "train.py", line 99, in main
engine.train(train_loader, model, optimizer, epoch, opt=options)
File "/home/jbadia/AMFMN/AMFMN/engine.py", line 52, in train
scores = model(input_visual, input_text, lengths)
File "/home/jbadia/anaconda3/envs/amfmn/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/jbadia/AMFMN/AMFMN/layers/AMFMN.py", line 44, in forward

Could it be a problem of versions?
Do you have a requirements file with exact versions?

Thank you very much

There are some blank lines in train_caps.txt

e.g. line 9631

The statistics based on the data in dataset_RSITMD.json do not match those in the paper. For example, the result of my statistics is 33 categories, but it is 32 categories in the paper, and there is no 'plane' category in the paper. The category with the highest number of images in my statistical results is 'storagestank', containing 251 images, while the category with the highest number of images in the thesis is 'industrial', containing 207 images.

this question came up during training：
[Errno 2] No such file or directory: '/seq2vec/dictionary.txt'

My path is set to：
dir_st: '/seq2vec/'
data_path: 'data/rsitmd_precomp/'
image_path: '/data/rsitmd_precomp/images/'

Also already downloaded ‘dictionary.txt’ ，Placed in a folder '/seq2vec/'

How do I need to fix this？

Each image input of the CAMP and SCAN models is 36 regional features. If you use ResNet18 for feature extraction, how to get these 36 regional features?

Hello blogger, I have the following problem while running the training code and would like to ask you the specific version information of torch installation