how to remove FLICKER effect in the depth

[andy500]

is there a way to reduce the flickering effect in the depth map

[mathmax12]

I am thinking the same issue. Do you get any idea? @ranftlr

[andy500]

thanks for your message l am using the version 3 and it seems to reduce the flicker effect a little bit have to say a very good way to create depth maps thanks to all of your team for the work you have done and making this a easy code to use keep the work going ☺

On Mon, 10 May 2021 at 07:18, mathmax12 @.***> wrote:

I am thinking the same issue. Do you get any idea? @ranftlr https://github.com/ranftlr

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/intel-isl/MiDaS/issues/87#issuecomment-836234084, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3NUI4U2XXF2WRXJ4UTTM53EBANCNFSM4W4VEXRA .

[semel1]

The same problem with flickering when processing image sequence. It caused by producing inconsistent depth map. Such" collateral effect " makes calculated depth maps unusable in case of processing image sequence from video. What could be done to reduce (ideally completely illuminate) such effect? May be "TecoGan"?

[andy500]

I agree with you. will have to see if they can fix it

On Mon, 14 Jun 2021 at 20:15, semel1 @.***> wrote:

The same problem with flickering when processing image sequence. It caused by producing inconsistent (in some cases) depth map. Such" collateral effect " makes calculated depth maps unusable in case of processing image sequence from video. What could be done to reduce (ideally completely illuminate) such effect? May be "TecoGan"?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/intel-isl/MiDaS/issues/87#issuecomment-860928695, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3KWSJYUBSASOUHDZDLTSZIMPANCNFSM4W4VEXRA .

[semel1]

It looks like solution here https://robust-cvd.github.io/. Waiting for code.

[andy500]

Thanks, I'll check it out.

On Thu, 24 Jun 2021 at 15:41, semel1 @.***> wrote:

It looks like solution here https://robust-cvd.github.io/. Waiting for code.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/intel-isl/MiDaS/issues/87#issuecomment-867692001, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3KUCHP4RIA37UHQHY3TUM7YLANCNFSM4W4VEXRA .

[3dsf]

@semel1 looks like the code is up now :)

[andy500]

so try the code on colab it was about 20 minutes to do on a video of 150 frames very slow

On Sat, 26 Jun 2021 at 20:33, 3dsf @.***> wrote:

@semel1 https://github.com/semel1 looks like the code is up now :)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/intel-isl/MiDaS/issues/87#issuecomment-869051491, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3OR42OWZO3GD3PX34DTUYTSNANCNFSM4W4VEXRA .

[andriilitvynchuk]

Midas has a flickering effect because original training has no color augmentations. To remove that authors should retrain the model with it or you need to reproduce training by yourself with color augmentation. If it is not the case, you can search on blind temporal consistency task, some solutions from that field can probably help you

[andy500]

thanks for the reply are you going to do a model with color augmentations training and why was the model not done with color augmentations did you not see that with out color augmentations that your model would have a lot of flickering in the frames l am using dpt which has not a lot of flickering thanks

On Fri, 3 Dec 2021 at 13:11, Andrii Litvynchuk @.***> wrote:

Midas has a flickering effect because original training has no color augmentations. To remove that authors should retain the model with it or you need to reproduce training by yourself with color augmentation.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/isl-org/MiDaS/issues/87#issuecomment-985507453, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3OHBWPVFOFIWZ75SW3UPC6ZFANCNFSM4W4VEXRA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[andriilitvynchuk]

I am not the creator of Midas so I cannot answer your questions. Probably, authors will include color augmentations in future version of Midas

[jiaqixuac]

Hi @andy500 , did you find a good solution or alternative for this? I want to infer depth for some random videos. Thanks.

[andy500]

Thanks for the message no still have flickering am using DPT-main code model_type dpt_large does reduce the flickering effect l am hoping there will be a update to model that do not create the flickering effects thank you

On Thu, 9 Dec 2021 at 07:50, Jiaqi Xu @.***> wrote:

Hi @andy500 https://github.com/andy500 , did you find a good solution or alternative for this? I want to infer depth for some random videos. Thanks.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/isl-org/MiDaS/issues/87#issuecomment-989597315, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3LGQUSWIOQDJYJFXR3UQBNUPANCNFSM4W4VEXRA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[jiaqixuac]

hhh Thank you. I tried with DPT-large but still some flickering and harms the downstream task somehow. It seems this is a hard problem for the general-purpose consistent video depth estimation.

[andy500]

Let hope there will update with a new model we can only hope

On Sat, 11 Dec 2021 at 09:17, Jiaqi Xu @.***> wrote:

hhh Thank you. I tried with DPT-large but still some flickering and harms the downstream task somehow. It seems this is a hard problem for the general-purpose consistent video depth estimation.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/isl-org/MiDaS/issues/87#issuecomment-991566972, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3MZQK2DIL2YNLFHFM3UQMJJNANCNFSM4W4VEXRA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[dpredie]

@andy500 this one is really good and efficient https://github.com/yu-li/TCMonoDepth

[andy500]

try out get this error Creating model... Loading model from ./weights/_ckpt.pt.tar Loading model done... Traceback (most recent call last): File "demo.py", line 148, in run(args) File "demo.py", line 76, in run args.resize_size[0], #width TypeError: 'int' object is not subscriptable

On Sun, 13 Feb 2022 at 14:45, dpredie @.***> wrote:

@andy500 https://github.com/andy500 this one is really good and efficient https://github.com/yu-li/TCMonoDepth

— Reply to this email directly, view it on GitHub https://github.com/isl-org/MiDaS/issues/87#issuecomment-1038166891, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3ILMMD6WDOOOGDU2LLU267YLANCNFSM4W4VEXRA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You are receiving this because you were mentioned.Message ID: @.***>

[semel1]

Got the same error.