Would you please tell the torch seed of 2 pretrained models.

[sjy234sjy234]

Would you please tell the torch seed of 2 pretrained models.
I found the result varies with torch seed, and I ran many times but failed to realize 76.2 for caption retrieval. I only got 75.6 instead.

[ahagary]

In addition to adapting to the python or torch version, did you have modify the code?
I use the provided code, dowload the pretrain model and use:
python evaluation_models.py
but I have only got (i2t: R@1 66.9 t2i:R@1 49.4) on flickr30k_precomp.
would you please provided some advices? Thank you very much! @sjy234sjy234 @KunpengLi1994

[sjy234sjy234]

In addition to adapting to the python or torch version, did you have modify the code?
I use the provided code, dowload the pretrain model and use:
python evaluation_models.py
but I have only got (i2t: R@1 66.9 t2i:R@1 49.4) on flickr30k_precomp.
would you please provided some advices? Thank you very much! @sjy234sjy234 @KunpengLi1994

When using torch 1+, the result is less than torch 0.4.

[ahagary]

I have tried your advise last week.And I think I still should reply to you to express my gratitude for you.Your suggestion is very effective.I use PY2.7 and torch0.4.1, then I got the result R@1 71.5 R@1i 54.8 on f30k_precomp.It's incredible that the version has so much inpact on the results. @sjy234sjy234

[By-he]

Hello, look at you replicating the author's experiment. I would like to ask you about the problem of "MemoryError: Unable to allocate 31.1 GiB for an Array with shape (8352423936,) and data Type Float32" when running the code. May I ask which parameters can be modified to work normally? My computer configuration is RTX2060, 6G video memory and 16G running memory

[ahagary]

Hello, look at you replicating the author's experiment. I would like to ask you about the problem of "MemoryError: Unable to allocate 31.1 GiB for an Array with shape (8352423936,) and data Type Float32" when running the code. May I ask which parameters can be modified to work normally? My computer configuration is RTX2060, 6G video memory and 16G running memory

This is caused by insufficient CPU memory , You can try to divide train_ims. npy in coco_precomp into multiple files, and then change __get_items__ of PrecompDataset. Although the data reading time will increase, but you can run the code normally

[sjy234sjy234]

your gpu memory might not big enough. Try smaller batch size see if it is. 发自我的iPhone

…

------------------ Original ------------------ From: heyi <[email protected]&gt; Date: Mon,Sep 21,2020 9:13 AM To: KunpengLi1994/VSRN <[email protected]&gt; Cc: JiangyangShen(Sam) <[email protected]&gt;, Mention <[email protected]&gt; Subject: Re: [KunpengLi1994/VSRN] Would you please tell the torch seed of 2 pretrained models. (#17) Hello, look at you replicating the author's experiment. I would like to ask you about the problem of "MemoryError: Unable to allocate 31.1 GiB for an Array with shape (8352423936,) and data Type Float32" when running the code. May I ask which parameters can be modified to work normally? My computer configuration is RTX2060, 6G video memory and 16G running memory This is caused by insufficient CPU memory , You can try to divide train_ims. npy in coco_precomp into multiple files, and then change __get_items__ of PrecompDataset. Although the data reading time will increase, but you can run the code normally — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[HHeracles]

@sjy234sjy234 @KunpengLi1994 ,Hello!May I ask you some questions? when I run the command"evaluation.evalrank("../Pretrain/pretrain_model/flickr/model_fliker_1.pth.tar", data_path="data", split="test", fold5=False)".I load the data is "data/f30k_precomp/test_ims.npy".
First, why is (36,2048) the shape of image?What does 36 mean here?36 regions?The image is directly input graphy, which consists of "Rs_GCN".What part of the code completes the generation of Salient Image Regions from bottomup attention?
Second,If I input the image is .jpg, how do I convert it to the desired form (36,2048) to meet the code input？