Yurima

188 posts

Yurima

Yurima

@xYurimax

Katılım Mayıs 2012
351 Takip Edilen48 Takipçiler
Yurima
Yurima@xYurimax·
@carlosedubarret that's pretty cool idea to pick the best from each method and combine them , great work 👏
English
0
0
1
9
Carlos Barreto
Carlos Barreto@carlosedubarret·
@xYurimax Actually its gem x. Im trying to integrate the global position from it to sam 3d.
English
1
0
1
57
Michael Black
Michael Black@Michael_J_Black·
I’ve been capturing 3D human motion for 30 years and today is maybe the biggest day in that history. We are presenting MAMMA at CVPR (oral session 2A). MAMMA is a markerless multi-camera system that has accuracy similar to marker-based systems.
GIF
English
55
255
2.1K
237K
Carlos Barreto
Carlos Barreto@carlosedubarret·
Thanks for sharing. It looks amazing, but the main 2 things are a no go for me. SMPL (since it requires an expensive license for commercial use) And the multicamera need. the commercial part is just a preparation for the future, in case I gotta a comercial work I wouldnt like to have to search another solution to work, and for the price of the license its something I wouldnt be able to pay. The multicamera thing, is more related to when I started trying markerless mocap solutions. Spent a few years trying multicamera solutions, and even when easymocap was in its start, the results were already very good, but the setup and all the trouble to set multple cameras.... Ended up not for me 😅 At least for the work that I'm trying to do, but thanks a lot for taging me. 💪
English
1
0
1
22
Carlos Barreto
Carlos Barreto@carlosedubarret·
CEB SAM 3d Body Loader (with sound 🔈) Footage from Anastasia Sereda on youtube Process: 1- selecting frames on the 60fps footage took about 15 minutes (287 frames were selected) 2- Processing using SAM 3d Body took about 20 minutes 3- loaded in blender #mocap #b3d
Carlos Barreto@carlosedubarret

CEB SAM 3d Body loader RELEASED 🙌 tutorial attached More information, links to download (for supporters) and youtube tutorial at: patreon.com/posts/ceb-sam-…

English
9
98
1.1K
159.1K
Yurima
Yurima@xYurimax·
@carlosedubarret hmm interesting maybe the scaling is the issue,i will test more to see if there is a permanent fix for this ,thanks mate 🙏
English
0
0
1
22
Carlos Barreto
Carlos Barreto@carlosedubarret·
@xYurimax Yeah I tried stable motion. I fell (but not sure) that the sliding might be caused by a diferent scale between the original mesh and the one being retarget to. Gotta test it.
English
1
0
1
37
Yurima
Yurima@xYurimax·
@carlosedubarret i tried so many repos to deal with foot sliding ,there is another repo from the same author i guess you tried it before it's called stable motion , i used it to fix foot sliding and the visualization looked so nice but when converting data it introduced the same issue 😂
English
1
0
1
39
Carlos Barreto
Carlos Barreto@carlosedubarret·
interesting, I didnt stop to notice that issue. I mean, I noticed that when bringind the dta to blender it had some foot problems, but I didnt noticed that the original data didnt. This is the first time I got the data in blender, after some more work it might get better. thanks for pointing it out.
English
1
0
1
67
Yurima
Yurima@xYurimax·
@DragonGroky @spiritbuun jpst.it/4_fc6 add this text at the end of your prompt "only output code , no other text" , the speed varies also its not consistent with longer context ,speed drops with time, the speed you get is so random ,you may get +150 t/s or 80 t/s with the same flags, enjoy
English
0
0
1
33
buun
buun@spiritbuun·
Now that the dust has settled, is DFlash real or hype? Today I was able to break the 200 tok/s barrier with 27B on a single RTX 3090: run 1: 194.6 t/s, accept=155/180 run 2: 205.1 t/s, accept=155/180 run 3: 206.3 t/s, accept=155/180 But what about real world usage? Well,
buun tweet media
English
21
9
212
33.6K
Yurima
Yurima@xYurimax·
@fubyo nice ,does the speed stay consistent or degrade ,cause it drops with time for me , also does it transfer to harnesses or only in webui
English
0
0
0
29
Dio Fu
Dio Fu@fubyo·
@xYurimax @spiritbuun i am on Windows 10, similar args. also same performance :-S
Rafina, Greece 🇬🇷 English
1
0
2
39
Yurima
Yurima@xYurimax·
@spiritbuun all good i will test it more if i found any solution i will put it here , thanks again mate 🙏
English
0
0
1
20
buun
buun@spiritbuun·
your harness should still be using the llama-server's API in the backend, right? Makes no sense that it wouldn't work without system prompt. I'll look into it and see if I can reproduce with claude code. Harness isn't open source, dw it's not as a good or feature complete as cc/kilo
English
1
0
1
70
Yurima
Yurima@xYurimax·
@spiritbuun tbf i have no clue i tried with no system prompt but still same issue i guess that's because the harness points to only qwen 3.6 27b not like llamacpp which combine the draft model with the main model, is your harness open source ?
English
1
0
0
54
buun
buun@spiritbuun·
@xYurimax I use my own harness but yeah I am seeing similar reports about hermes. Maybe the result of a big system prompt?
English
1
0
1
49