Flowframes Early Access Download: 1.37.0
🕑 Added 2022-04-17 18:23:20 +0000 UTCSorry for the long wait - I've been rather busy with Nmkoder and some IRL stuff.
Anyway, here it is - Flowframes 1.37, now with RIFE 4.0 on NCNN (for AMD/Intel GPUs).
Not only has it been added to NCNN, but it's also a lot faster, making RIFE-NCNN roughly as fast as the CUDA version.
With this new model, it's now also possible to interpolate with ANY factor, not just 4x/8x.
NCNN (unlike CUDA) even supports fractional factors, e.g. 2.5x to get from 24 to 60 FPS. However, scene detection does not work yet for fractional factors, this is something I'm still working on.
Full changelog on Discord and in Flowframes.
Comments
JackalBlender
Be careful, the server for the installation of Flowframes has an expired TLS certificate, as such the installer doesn't work anymore. What's worse, when I try to use the old version, it got deleted by the installer ...
Kenneth Flournoy
great work appreciate you keeping to donate to you yes please vodka for you and me
Cometakah
I don't speak English, because i'm from Chile (South America) But i'm need say you: Thanks! This program Helps me a LOT in many things that i do, Seriously Thanks Man, I love your work.
Chris Liu
Feature request: Prevent sleep or hibernate when running interpolation.
Chris Liu
Bug report: If I set output folder to a network share, application will report " Not enough disk space on '\\/' (0.0 GB)! (Message)". But actually, my share folder has 7TB free space. Please fix this, thank you very much!
Jens Moths
Great work! I get weird artifacts with RIFE4.0, things move around in the center of the picture.
User
[4453] [4-19-2022 13:0:50]: [E] Traceback (most recent call last): [4454] [4-19-2022 13:0:50]: [E] File "main.py", line 425, in [4455] [4-19-2022 13:0:50]: [E] main() [4456] [4-19-2022 13:0:50]: [E] File "main.py", line 166, in main [4457] [4-19-2022 13:0:50]: [E] final_pred_save_path = test(final_test_loader, model_net, [4458] [4-19-2022 13:0:50]: [E] File "main.py", line 352, in test [4459] [4-19-2022 13:0:50]: [E] for testIndex, (frames, t_value, scene_name, frameRange) in enumerate(test_loader): [4460] [4-19-2022 13:0:50]: [E] File "C:\Users\User\AppData\Local\Flowframes\FlowframesData\pkgs\py-amp\Lib\site-packages\torch\utils\data\dataloader.py", line 435, in __next__ [4461] [4-19-2022 13:0:50]: [E] data = self._next_data() [4462] [4-19-2022 13:0:50]: [E] File "C:\Users\User\AppData\Local\Flowframes\FlowframesData\pkgs\py-amp\Lib\site-packages\torch\utils\data\dataloader.py", line 475, in _next_data [4463] [4-19-2022 13:0:50]: [E] data = self._dataset_fetcher.fetch(index) # may raise StopIteration [4464] [4-19-2022 13:0:50]: [E] File "C:\Users\User\AppData\Local\Flowframes\FlowframesData\pkgs\py-amp\Lib\site-packages\torch\utils\data\_utils\fetch.py", line 44, in fetch [4465] [4-19-2022 13:0:50]: [E] data = [self.dataset[idx] for idx in possibly_batched_index] [4466] [4-19-2022 13:0:50]: [E] File "C:\Users\User\AppData\Local\Flowframes\FlowframesData\pkgs\py-amp\Lib\site-packages\torch\utils\data\_utils\fetch.py", line 44, in [4467] [4-19-2022 13:0:50]: [E] data = [self.dataset[idx] for idx in possibly_batched_index] [4468] [4-19-2022 13:0:50]: [E] File "C:\Users\User\AppData\Local\Flowframes\FlowframesData\pkgs\xvfi-cuda\utils.py", line 300, in __getitem__ [4469] [4-19-2022 13:0:50]: [E] frames = frames_loader_test(self.args, I0I1It_Path, None) [4470] [4-19-2022 13:0:50]: [E] File "C:\Users\User\AppData\Local\Flowframes\FlowframesData\pkgs\xvfi-cuda\utils.py", line 154, in frames_loader_test [4471] [4-19-2022 13:0:50]: [E] (ih, iw, c) = frame.shape [4472] [4-19-2022 13:0:50]: [E] AttributeError: 'NoneType' object has no attribute 'shape'
User
Can't get XVFI to work at all. All other interpolation option work fine. RTX 3090.
Dragan Petronijević
RIFE 4.0 is so good for old GPUs like HD7970. It finally can interpolate 1080p video without crashing & at much faster rate. Thanks a lot!
Frustumalignedvoxelfog
Thanks man!