-
Notifications
You must be signed in to change notification settings - Fork 461
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Does the video need to be cropped? #119
Comments
Hello, do you have any progress now? I have the same need. Could we discuss it? |
In principle the model should handle any resolution since it is
convolutional.
Joao
…On Sat, 28 Jan 2023, 08:17 Coolnerdn, ***@***.***> wrote:
Hello, do you have any progress now? I have the same need. Could we
discuss it?
—
Reply to this email directly, view it on GitHub
<#119 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
Thank you for the response. I want to use my own dataset to fine-tune the I3D network. Is there any requirement on the number of frames? Must it be the number of frames during training?
哈哈
***@***.***
…------------------ 原始邮件 ------------------
发件人: "deepmind/kinetics-i3d" ***@***.***>;
发送时间: 2023年1月28日(星期六) 晚上7:15
***@***.***>;
***@***.******@***.***>;
主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped? (Issue #119)
In principle the model should handle any resolution since it is
convolutional.
Joao
On Sat, 28 Jan 2023, 08:17 Coolnerdn, ***@***.***> wrote:
> Hello, do you have any progress now? I have the same need. Could we
> discuss it?
>
> —
> Reply to this email directly, view it on GitHub
> <#119 (comment);,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
> .
> You are receiving this because you are subscribed to this thread.Message
> ID: ***@***.***>
>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Thank you for the response. I want to use my own dataset to fine-tune the I3D network. Is there any requirement on the number of frames? Must it be the number of frames during training? |
No, you can use a different number of frames. It may be that it will not
work as well if you use fewer frames, but you'd have to try to be sure.
Best,
Joao
…On Sun, Jan 29, 2023 at 4:41 AM Coolnerdn ***@***.***> wrote:
In principle the model should handle any resolution since it is
convolutional. Joao
… <#m_-454727744888391567_>
On Sat, 28 Jan 2023, 08:17 Coolnerdn, *@*.*> wrote: Hello, do you have
any progress now? I have the same need. Could we discuss it? — Reply to
this email directly, view it on GitHub <#119 (comment)
<#119 (comment)>>,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
<https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
. You are receiving this because you are subscribed to this thread.Message
ID: @.*>
Thank you for the response. I want to use my own dataset to fine-tune the
I3D network. Is there any requirement on the number of frames? Must it be
the number of frames during training?
—
Reply to this email directly, view it on GitHub
<#119 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ADXKU2RISFEA5JAWPB3RQHTWUXYHBANCNFSM5KSWB4KA>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
I see. There is one more question. Did you use the same number of frames when you train the network? The video lengths I use to fine-tune are not the same, so the number of frames are different and can't be fed directly into the network. How did you solve this problem? Or did you crop the video to the same length during training?
哈哈
***@***.***
…------------------ 原始邮件 ------------------
发件人: "deepmind/kinetics-i3d" ***@***.***>;
发送时间: 2023年1月29日(星期天) 下午2:22
***@***.***>;
***@***.******@***.***>;
主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped? (Issue #119)
No, you can use a different number of frames. It may be that it will not
work as well if you use fewer frames, but you'd have to try to be sure.
Best,
Joao
On Sun, Jan 29, 2023 at 4:41 AM Coolnerdn ***@***.***> wrote:
> In principle the model should handle any resolution since it is
> convolutional. Joao
> … <#m_-454727744888391567_>
> On Sat, 28 Jan 2023, 08:17 Coolnerdn, *@*.*> wrote: Hello, do you have
> any progress now? I have the same need. Could we discuss it? — Reply to
> this email directly, view it on GitHub <#119 (comment)
> <#119 (comment);,
> or unsubscribe
> https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
> <https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
> . You are receiving this because you are subscribed to this thread.Message
> ID: @.*>
>
> Thank you for the response. I want to use my own dataset to fine-tune the
> I3D network. Is there any requirement on the number of frames? Must it be
> the number of frames during training?
>
> —
> Reply to this email directly, view it on GitHub
> <#119 (comment);,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/ADXKU2RISFEA5JAWPB3RQHTWUXYHBANCNFSM5KSWB4KA>
> .
> You are receiving this because you commented.Message ID:
> ***@***.***>
>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
it should be possible to feed an arbitrary number of frames to the network,
what is the error you get ?
You should just have to replace the 64 constant in places like this by your
number of frames:
inp = tf.placeholder(tf.float32, [None, 64, _IMAGE_SIZE, _IMAGE_SIZE, 3])
Joao
…On Sun, Jan 29, 2023 at 8:37 AM Coolnerdn ***@***.***> wrote:
I see. There is one more question. Did you use the same number
of frames when you train the network? The video lengths I use to fine-tune
are not the same, so the number of frames are different and can't be fed
directly into the network. How did you solve this problem? Or
did you crop the video to the same length during training?
哈哈
***@***.***
------------------ 原始邮件 ------------------
发件人: "deepmind/kinetics-i3d" ***@***.***>;
发送时间: 2023年1月29日(星期天) 下午2:22
***@***.***>;
***@***.******@***.***>;
主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped?
(Issue #119)
No, you can use a different number of frames. It may be that it will not
work as well if you use fewer frames, but you'd have to try to be sure.
Best,
Joao
On Sun, Jan 29, 2023 at 4:41 AM Coolnerdn ***@***.***> wrote:
> In principle the model should handle any resolution since it is
> convolutional. Joao
> … <#m_-454727744888391567_>
> On Sat, 28 Jan 2023, 08:17 Coolnerdn, *@*.*> wrote: Hello, do you
have
> any progress now? I have the same need. Could we discuss it? — Reply
to
> this email directly, view it on GitHub <#119 (comment)
> <
#119 (comment);,
> or unsubscribe
>
https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
> <
https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
> . You are receiving this because you are subscribed to this
thread.Message
> ID: @.*>
>
> Thank you for the response. I want to use my own dataset to fine-tune
the
> I3D network. Is there any requirement on the number of frames? Must
it be
> the number of frames during training?
>
> —
> Reply to this email directly, view it on GitHub
> <
#119 (comment);,
> or unsubscribe
> <
https://github.com/notifications/unsubscribe-auth/ADXKU2RISFEA5JAWPB3RQHTWUXYHBANCNFSM5KSWB4KA>
> .
> You are receiving this because you commented.Message ID:
> ***@***.***>
>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub
<#119 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ADXKU2S3FZ4O6QXZWXQZZEDWUYT2ZANCNFSM5KSWB4KA>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
I mean, I'm not sure how many frames I'm going to input, so I can't replace 64 with a specific number, because the number of frames could be any integer from 20 to 90. Can this data be used directly to train networks? |
Ok got it. People typically pad the videos to a fixed size (or slice it
into chunks if too large so does not fit in memory).
Joao
…On Sun, 29 Jan 2023, 09:49 Coolnerdn, ***@***.***> wrote:
it should be possible to feed an arbitrary number of frames to the
network, what is the error you get ? You should just have to replace the 64
constant in places like this by your number of frames: inp =
tf.placeholder(tf.float32, [None, 64,
*IMAGE_SIZE, IMAGE_SIZE, 3]) Joao … <#m_-6833791440840358865_> On Sun, Jan
29, 2023 at 8:37 AM Coolnerdn @.> wrote: I see. There is one more
question. Did you use the same number of frames when you train the
network? The video lengths I use to fine-tune are not the same, so the
number of frames are different and can't be fed directly into the network.
How did you solve this problem? Or did you crop the video to the same
length during training? 哈哈 @. ------------------ 原始邮件 ------------------
发件人: "deepmind/kinetics-i3d" @.>; 发送时间: 2023年1月29日(星期天) 下午2:22 @.>; @.@.>;
主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped? (Issue
#119 <#119>) No, you can use
a different number of frames. It may be that it will not work as well if
you use fewer frames, but you'd have to try to be sure. Best, Joao On Sun,
Jan 29, 2023 at 4:41 AM Coolnerdn @.***> wrote: > In principle the model
should handle any resolution since it is > convolutional. Joao > …
<#m-454727744888391567*> > On Sat, 28 Jan 2023, 08:17 Coolnerdn, *@*.*>
wrote: Hello, do you have > any progress now? I have the same need. Could
we discuss it? — Reply to > this email directly, view it on GitHub <#119
<#119> (comment) > < #119
(comment)
<#119 (comment)>>>,
> or unsubscribe >
https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
<https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
> <
https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
<https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>>
> . You are receiving this because you are subscribed to this
thread.Message > ID: @.*> > > Thank you for the response. I want to use
my own dataset to fine-tune the > I3D network. Is there any requirement on
the number of frames? Must it be > the number of frames during training? >
> — > Reply to this email directly, view it on GitHub > < #119 (comment)
<#119 (comment)>>,
> or unsubscribe > <
https://github.com/notifications/unsubscribe-auth/ADXKU2RISFEA5JAWPB3RQHTWUXYHBANCNFSM5KSWB4KA>
> . > You are receiving this because you commented.Message ID: > *@*.*> >
— Reply to this email directly, view it on GitHub, or unsubscribe. You are
receiving this because you commented.Message ID: @.*> — Reply to this
email directly, view it on GitHub <#119 (comment)
<#119 (comment)>>,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ADXKU2S3FZ4O6QXZWXQZZEDWUYT2ZANCNFSM5KSWB4KA
. You are receiving this because you commented.Message ID: *@*.***>
I mean, I'm not sure how many frames I'm going to input, so I can't
replace 64 with a specific number, because the number of frames could be
any integer from 20 to 90. Can this data be used directly to train networks?
—
Reply to this email directly, view it on GitHub
<#119 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ADXKU2VTA2R4DECJC2EJ2CLWUY4MLANCNFSM5KSWB4KA>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
I see. Thank you for your kind reply.
Best wishes!
哈哈
***@***.***
…------------------ 原始邮件 ------------------
发件人: "deepmind/kinetics-i3d" ***@***.***>;
发送时间: 2023年1月29日(星期天) 下午5:54
***@***.***>;
***@***.******@***.***>;
主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped? (Issue #119)
Ok got it. People typically pad the videos to a fixed size (or slice it
into chunks if too large so does not fit in memory).
Joao
On Sun, 29 Jan 2023, 09:49 Coolnerdn, ***@***.***> wrote:
> it should be possible to feed an arbitrary number of frames to the
> network, what is the error you get ? You should just have to replace the 64
> constant in places like this by your number of frames: inp =
> tf.placeholder(tf.float32, [None, 64,
>
> *IMAGE_SIZE, IMAGE_SIZE, 3]) Joao … <#m_-6833791440840358865_> On Sun, Jan
> 29, 2023 at 8:37 AM Coolnerdn @.> wrote: I see. There is one more
> question. Did you use the same number of frames when you train the
> network? The video lengths I use to fine-tune are not the same, so the
> number of frames are different and can't be fed directly into the network.
> How did you solve this problem? Or did you crop the video to the same
> length during training? 哈哈 @. ------------------ 原始邮件 ------------------
> 发件人: "deepmind/kinetics-i3d" @.>; 发送时间: 2023年1月29日(星期天) 下午2:22 @.>; @.@.>;
> 主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped? (Issue
> #119 <#119;) No, you can use
> a different number of frames. It may be that it will not work as well if
> you use fewer frames, but you'd have to try to be sure. Best, Joao On Sun,
> Jan 29, 2023 at 4:41 AM Coolnerdn @.***> wrote: > In principle the model
> should handle any resolution since it is > convolutional. Joao > …
> <#m-454727744888391567*> > On Sat, 28 Jan 2023, 08:17 Coolnerdn, *@*.*>
> wrote: Hello, do you have > any progress now? I have the same need. Could
> we discuss it? — Reply to > this email directly, view it on GitHub <#119
> <#119; (comment) > < #119
> (comment)
> <#119 (comment);,
> > or unsubscribe >
> https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
> <https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
> > <
> https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
> <https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>>
> > . You are receiving this because you are subscribed to this
> thread.Message > ID: @.*> > > Thank you for the response. I want to use
> my own dataset to fine-tune the > I3D network. Is there any requirement on
> the number of frames? Must it be > the number of frames during training? >
> > — > Reply to this email directly, view it on GitHub > < #119 (comment)
> <#119 (comment);,
> > or unsubscribe > <
> https://github.com/notifications/unsubscribe-auth/ADXKU2RISFEA5JAWPB3RQHTWUXYHBANCNFSM5KSWB4KA>
> > . > You are receiving this because you commented.Message ID: > *@*.*> >
> — Reply to this email directly, view it on GitHub, or unsubscribe. You are
> receiving this because you commented.Message ID: @.*> — Reply to this
> email directly, view it on GitHub <#119 (comment)
> <#119 (comment);,
> or unsubscribe
> https://github.com/notifications/unsubscribe-auth/ADXKU2S3FZ4O6QXZWXQZZEDWUYT2ZANCNFSM5KSWB4KA
> . You are receiving this because you commented.Message ID: *@*.***>
>
> I mean, I'm not sure how many frames I'm going to input, so I can't
> replace 64 with a specific number, because the number of frames could be
> any integer from 20 to 90. Can this data be used directly to train networks?
>
> —
> Reply to this email directly, view it on GitHub
> <#119 (comment);,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/ADXKU2VTA2R4DECJC2EJ2CLWUY4MLANCNFSM5KSWB4KA>
> .
> You are receiving this because you commented.Message ID:
> ***@***.***>
>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Could you please tell me how to pad the video to a fixed size in the I3D paper? Can I add it before and after a shorter video to achieve this goal, and will this affect the recognition of the model?
I'm currently considering of using your work for cross-modality tasks, but since I've never been involved in CV, I really need your advice. Thanks again.
哈哈
***@***.***
…------------------ 原始邮件 ------------------
发件人: "deepmind/kinetics-i3d" ***@***.***>;
发送时间: 2023年1月29日(星期天) 下午5:54
***@***.***>;
***@***.******@***.***>;
主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped? (Issue #119)
Ok got it. People typically pad the videos to a fixed size (or slice it
into chunks if too large so does not fit in memory).
Joao
On Sun, 29 Jan 2023, 09:49 Coolnerdn, ***@***.***> wrote:
> it should be possible to feed an arbitrary number of frames to the
> network, what is the error you get ? You should just have to replace the 64
> constant in places like this by your number of frames: inp =
> tf.placeholder(tf.float32, [None, 64,
>
> *IMAGE_SIZE, IMAGE_SIZE, 3]) Joao … <#m_-6833791440840358865_> On Sun, Jan
> 29, 2023 at 8:37 AM Coolnerdn @.> wrote: I see. There is one more
> question. Did you use the same number of frames when you train the
> network? The video lengths I use to fine-tune are not the same, so the
> number of frames are different and can't be fed directly into the network.
> How did you solve this problem? Or did you crop the video to the same
> length during training? 哈哈 @. ------------------ 原始邮件 ------------------
> 发件人: "deepmind/kinetics-i3d" @.>; 发送时间: 2023年1月29日(星期天) 下午2:22 @.>; @.@.>;
> 主题: Re: [deepmind/kinetics-i3d] Does the video need to be cropped? (Issue
> #119 <#119;) No, you can use
> a different number of frames. It may be that it will not work as well if
> you use fewer frames, but you'd have to try to be sure. Best, Joao On Sun,
> Jan 29, 2023 at 4:41 AM Coolnerdn @.***> wrote: > In principle the model
> should handle any resolution since it is > convolutional. Joao > …
> <#m-454727744888391567*> > On Sat, 28 Jan 2023, 08:17 Coolnerdn, *@*.*>
> wrote: Hello, do you have > any progress now? I have the same need. Could
> we discuss it? — Reply to > this email directly, view it on GitHub <#119
> <#119; (comment) > < #119
> (comment)
> <#119 (comment);,
> > or unsubscribe >
> https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
> <https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>
> > <
> https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA
> <https://github.com/notifications/unsubscribe-auth/ADXKU2UZYJUGAFBWDQWCX3DWUTIZFANCNFSM5KSWB4KA>>
> > . You are receiving this because you are subscribed to this
> thread.Message > ID: @.*> > > Thank you for the response. I want to use
> my own dataset to fine-tune the > I3D network. Is there any requirement on
> the number of frames? Must it be > the number of frames during training? >
> > — > Reply to this email directly, view it on GitHub > < #119 (comment)
> <#119 (comment);,
> > or unsubscribe > <
> https://github.com/notifications/unsubscribe-auth/ADXKU2RISFEA5JAWPB3RQHTWUXYHBANCNFSM5KSWB4KA>
> > . > You are receiving this because you commented.Message ID: > *@*.*> >
> — Reply to this email directly, view it on GitHub, or unsubscribe. You are
> receiving this because you commented.Message ID: @.*> — Reply to this
> email directly, view it on GitHub <#119 (comment)
> <#119 (comment);,
> or unsubscribe
> https://github.com/notifications/unsubscribe-auth/ADXKU2S3FZ4O6QXZWXQZZEDWUYT2ZANCNFSM5KSWB4KA
> . You are receiving this because you commented.Message ID: *@*.***>
>
> I mean, I'm not sure how many frames I'm going to input, so I can't
> replace 64 with a specific number, because the number of frames could be
> any integer from 20 to 90. Can this data be used directly to train networks?
>
> —
> Reply to this email directly, view it on GitHub
> <#119 (comment);,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/ADXKU2VTA2R4DECJC2EJ2CLWUY4MLANCNFSM5KSWB4KA>
> .
> You are receiving this because you commented.Message ID:
> ***@***.***>
>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Can the model handle not square input of the video? Or does it need to be 224x224?
I plan to fine-tune it on another dataset so will be removing the top layer.
The text was updated successfully, but these errors were encountered: