train process about siamese-rpn HOT 8 CLOSED

There is a new version now in 59version folder. You can train with a pretrained model on imagenet with it without warm up.

from siamese-rpn.

There is a new version now in 59version folder. You can train with a pretrained model on imagenet with it without warm up.

can you provide some main tricks during your tuning? by the way, youtube-bb only bring 1 auc gain?

from siamese-rpn.

There is a new version now in 59version folder. You can train with a pretrained model on imagenet with it without warm up.
i have tried the 59version without youtub-bb and change pairs_per_video_per_epoch = 10, during 50 epoch training, the auc on otb100 is only about 0.5, so how to modify the config file when we train the model with only VID dataset?

from siamese-rpn.

from siamese-rpn.

0.51auc on otb100 without YouTube dataset

but i see your 54version can get 0.54 on otb50 without youtube-bb, but with same model which gets 0.5 on otb100, the performance on otb50 is only about 0.45, is this result correct? generally, AUC result on otb50 is lower than otb100, so what is the correct config on OTB50 ?

from siamese-rpn.

from siamese-rpn.

I can only get 0.51auc on otb100 without YouTube dataset。------------------ Original ------------------From: ucasqcz [email protected]Date: Tue,Apr 2,2019 2:26 PMTo: HelloRicky123/Siamese-RPN [email protected]Cc: HelloRicky123 [email protected], Comment [email protected]Subject: Re: [HelloRicky123/Siamese-RPN] train process (#15) There is a new version now in 59version folder. You can train with a pretrained model on imagenet with it without warm up. i have tried the 59version without youtub-bb and change pairs_per_video_per_epoch = 10, during 50 epoch training, the auc on otb100 is only about 0.5, so how to modify the config file when we train the model with only VID dataset? —You are receiving this because you commented.Reply to this email directly, view it on GitHub, or mute the thread. {"api_version":"1.0","publisher":{"api_key":"05dde50f1d1a384dd78767c55493e4bb","name":"GitHub"},"entity":{"external_key":"github/HelloRicky123/Siamese-RPN","title":"HelloRicky123/Siamese-RPN","subtitle":"GitHub repository","main_image_url":"https://github.githubassets.com/images/email/message_cards/header.png","avatar_image_url":"https://github.githubassets.com/images/email/message_cards/avatar.png","action":{"name":"Open in GitHub","url":"[https://github.com/HelloRicky123/Siamese-RPN"}},"updates":{"snippets":{"icon":"PERSON","message":"@ucasqcz in #15: \u003e There is a new version now in 59version folder. You can train with a pretrained model on imagenet with it without warm up.\r\ni have tried the 59version without youtub-bb and change pairs_per_video_per_epoch = 10, during 50 epoch training, the auc on otb100 is only about 0.5, so how to modify the config file when we train the model with only VID dataset?\r\n"}],"action":{"name":"View Issue","url":"#15 (comment)"}}} [ { "@context": "http://schema.org", "@type": "EmailMessage", "potentialAction": { "@type": "ViewAction", "target": "#15 (comment)", "url": "#15 (comment)", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { "@type": "Organization", "name": "GitHub", "url": "https://github.com" } } ]
yes, but the 54version of your code also train the model without youtube, and the auc
on OTB50 is nearly 10 points higher on the 59version, I am confused about your results, which is the test dataset for the 0.54 auc performance?

from siamese-rpn.

I format this repo and you can try it.

from siamese-rpn.