Dain-App 0.43
Added 2020-09-22 14:22:09 +0000 UTCMirror: MediaFire
Google Drive block the link after a few downloads, so use the MediaFire mirror in this case.
What is in 0.43?
- Batch rendering should be working correctly in this version.
- Various minor bugs fixed
- You can now select the model you want to use on "Beta" tab
- I uploaded my test models in this build, I did find that they tend to produce worse results than the default model (best.pth), but perhaps someone can find some user-case that they perform better.
- Because of the extra models, this build is larger
What is wrong in 0.43?
- CLI is currently broken because of a few mistakes on my part. Should be fixed in the next few days.
- Mode 3 is currently not working because the changes to make batch work.
- Experimental is not working, in the next few weeks I will update the main algorithm to support the experimental behavior.
- There was a few reports of the app not working on the new 30X0 cards, will try to update a few libraries in the next build to see if fix it.
- Scene change seen to be ignoring all the batch interpolation if is detected as true.
What about Multi-GPU?
- I gave a few tries to make Multi-GPU work, but developing it without having two GPU's to debug it proved to be to difficult, even with a few friends that could test the build for me. So for now this is paused while I work on other features on the app.
Comments
Wow, batches are a real game-changer. I'm able to get my processing times to under a third of the original time. Tested using a 320x240 11s video on a i7-2700k 1070Ti system: batch=1 169s batch=2 83s batch=3 80s batch=4 83s batch=5 75s batch=6 68s batch=10 56s batch=12 53s batch=15 59s batch=18 49s batch=20 51s
Gareth Perks
2020-10-09 19:44:56 +0000 UTCResume Render is also broken it starts from the first frame instead of last processed frame. If you are rendering big animation and don't want to spend another 10 hours rendering it from start use 0.41 version and in config.json add "useMultiProcess": 0, property. Then just resume render and it should pickup where it left off last time.
2020-10-04 14:01:43 +0000 UTCHopeful the speed can be improved in the next versions, there still a lot of room for improvement.
GRisk
2020-10-02 12:11:33 +0000 UTCBatching gives a lot less speedup than I thought. Same video, 354 iterations, 640x480 input renders in 10:28 with a batch size of 3 and in 11:37 without batching. GPU is GTX1060 with 6 gigs of RAM.
naorunaoru
2020-10-01 22:20:51 +0000 UTCFP32 and the CUDA cores number is a big thing. I wonder how much faster will the 3080 be compared to my 1080ti.
Ka K Chan
2020-09-28 00:31:28 +0000 UTC
