Speed information request

[SarahSolaiman]

Hi,
I would like to thank you for the great work detailed in your paper.
Nevertheless, I was wondering about the execution speed of your method.
Could you please, give me some details concerning the inference time, your proposed method takes ?
I deeply appreciate your kind reply.

[sunnychencool]

Hi Sarah, Thank for your interest for our work. If you use our model in a real system (i.e. predict the pose frame-by-frame), the speed of the full model is about 4 times slower than the baseline TCN. However, for real system, you must use our 2D-3D model or TCN based on an image-2D model. Actually the speed of a 2D-3D model is typically far faster than an image-2D model. Therefore, the inference speed of the 2D-3D model is not the bottleneck. BTW, if you want to increase the speed for our model in real system. You can set the model to only predict the bone length by the first 100 or 200 frames and then fix the bone length. This will sharply increase the inference speed for the following frames since you don’t need to compute the bone length. We have tested this strategy and there isn’t obvious drop for the performance. Best wishes, Tianlang From: SarahSolaiman <notifications@github.com> Sent: 2020年4月23日 8:56 To: sunnychencool/Anatomy3D <Anatomy3D@noreply.github.com> Cc: Subscribed <subscribed@noreply.github.com> Subject: [sunnychencool/Anatomy3D] Speed information request (#1) Hi, I would like to thank you for the great work detailed in your paper. Nevertheless, I was wondering about the execution speed of your method. Could you please, give me some details concerning the inference time, your proposed method takes ? I deeply appreciate your kind reply. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_sunnychencool_Anatomy3D_issues_1&d=DwMCaQ&c=kbmfwr1Yojg42sGEpaQh5ofMHBeTl9EI2eaqQZhHbOU&r=rPOBJC5gKaoaPCPUTVTptPi9X8fSUVufMYyYR7i8ztA&m=ioR6cUJSFdPSuM7kQB_2xo-2s5-n24WY94YDB3bpFo0&s=38cVXSn1kVGIWoheiDdVvo-jZZqz2R-5Q65yu0INook&e=>, or unsubscribe<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AFQKPECWKDSWSDEKSP6O4LLROA3FBANCNFSM4MPBKLPQ&d=DwMCaQ&c=kbmfwr1Yojg42sGEpaQh5ofMHBeTl9EI2eaqQZhHbOU&r=rPOBJC5gKaoaPCPUTVTptPi9X8fSUVufMYyYR7i8ztA&m=ioR6cUJSFdPSuM7kQB_2xo-2s5-n24WY94YDB3bpFo0&s=0lcx8RLs9zcsg_K2HvM7Ubef_8bATyhV0E_9u4IPtOQ&e=>.

[SarahSolaiman]

Dear Dr. Tianlang,
My deepest thank's for your prompt reply.
I was wondering, and would greatly appreciate, if you can give some details and quantitative information related to the inference time on each used platform.
Sincere thanks for your cooperation.