Emsung
commented
1 year ago I've got the same issue.
Fresh install of Flowframes 1.40.0 `[00001114] [04-04-2023 21:34:43]: Failed to parse "Normal Speed" to int: De indeling van de invoertekenreeks is onjuist. [00001115] [04-04-2023 21:34:43]: RefreshExtensions(Both) - alpha = False pngOutput = False aviHqChroma = False proresHqChroma = False [00001116] [04-04-2023 21:34:43]: RefreshExtensions - Using '.jpg' for imported frames, using '.png' for interpolated frames
[00001119] [04-04-2023 21:34:43]: RefreshAlpha: model.supportsAlpha = False - outputSupportsAlpha = False - input ext: .png => alpha = False [00001120] [04-04-2023 21:34:43]: RefreshExtensions(Import) - alpha = False pngOutput = False aviHqChroma = False proresHqChroma = False [00001121] [04-04-2023 21:34:43]: RefreshExtensions - Using '.jpg' for imported frames, using '.png' for interpolated frames [00001122] [04-04-2023 21:34:43]: Getting media resolution (E:\stable-diffusion-webui\outputs\img2img-images\loopback-wave\4110793041) [00001123] [04-04-2023 21:34:43]: Cache contains this hash, using cached value. [00001124] [04-04-2023 21:34:43]: Un-rounded downscaled size: 512x768 [00001125] [04-04-2023 21:34:43]: Sequence compatible! [00001126] [04-04-2023 21:34:43]: [UI] Loading images from 4110793041... [00001127] [04-04-2023 21:34:43]: Symlink Import disabled, copying input frames... [00001128] [04-04-2023 21:34:44]: Processing frames... [00001129] [04-04-2023 21:34:44]: Downloading models... [00001130] [04-04-2023 21:34:44]: DownloadModelFiles(string ai = RIFE_NCNN_VS, string model = rife-v4.4, bool log = True) [00001131] [04-04-2023 21:34:44]: Computed CRC32 for 'flownet.bin' (9.8 MB): 2931561690 (11ms) [00001132] [04-04-2023 21:34:44]: Computed CRC32 for 'flownet.param' (13.6 KB): 2236004978 (0ms) [00001133] [04-04-2023 21:34:44]: Not Using AutoEnc: Using piped encoding. [00001134] [04-04-2023 21:34:44]: Getting media resolution (E:\stable-diffusion-webui\outputs\img2img-images\loopback-wave\4110793041) [00001135] [04-04-2023 21:34:44]: Cache contains this hash, using cached value. [00001136] [04-04-2023 21:34:44]: Un-rounded downscaled size: 512x768 [00001137] [04-04-2023 21:34:44]: [UI] Running RIFE (NCNN-VS)... [00001138] [04-04-2023 21:34:44]: Un-rounded downscaled size: 512x768 [00001139] [04-04-2023 21:34:44]: Un-rounded downscaled size: 512x768 [00001140] [04-04-2023 21:34:44]: Converted float 0 to Fraction 0/1 [00001145] [04-04-2023 21:34:44]: Converted float 1 to Fraction 1/1 [00001146] [04-04-2023 21:34:44]: Converted float 0 to Fraction 0/1 [00001147] [04-04-2023 21:34:44]: Getting media resolution (E:\stable-diffusion-webui\outputs\img2img-images\loopback-wave\4110793041) [00001148] [04-04-2023 21:34:44]: Cache contains this hash, using cached value. [00001149] [04-04-2023 21:34:44]: Un-rounded downscaled size: 512x768 [00001150] [04-04-2023 21:34:44]: Running AI... [00001151] [04-04-2023 21:34:45]: NCNN: Found GPU 0 with compute queue count 8 [00001153] [04-04-2023 21:34:45]: Starting GetProgressFromFfmpegLog() loop for log 'rife-ncnn-vs-log', target is 421 frames [00001154] [04-04-2023 21:34:45]: [UI] Interpolated 0/421 Frames (0%) - Average Speed: 0.00 FPS In / 0.00 FPS Out - Time: 05s - ETA: 08s [00001155] [04-04-2023 21:34:45]: Creating RIFE VS script. Model: rife-v4.4, Factor: 2, Res: 512x768, UHD: False, SC Sens: 0, GPU ID: 0, GPU Threads: 4, TTA: False, Loop: False, Match Duration: False, Dedupe: False, RT: False [00001156] [04-04-2023 21:34:45]: Un-rounded downscaled size: 512x768 [00001157] [04-04-2023 21:34:45]: Un-rounded downscaled size: 512x768 [00001158] [04-04-2023 21:34:45]: cmd.exe /C cd /D "E:\Flowframes\FlowframesData\pkgs\rife-ncnn-vs" & vspipe "E:\Flowframes\FlowframesData\pkgs\rife-ncnn-vs\rife.vpy" -c y4m - | "E:\Flowframes\FlowframesData\pkgs\av\ffmpeg" -y -r 60/1 -i pipe: -filter_complex [0:v]pad=width=ceil(iw/2)2:height=ceil(ih/2)2:color=black@0[vf] -map [vf] -movflags +faststart -c:v libx264 -g 600 -pix_fmt yuv420p -crf 16 -preset fast "E:\stable-diffusion-webui\outputs\img2img-images\loopback-wave\4110793041-2x-RIFE-RIFE4.4-60fps.mp4" [00001193] [04-04-2023 21:34:48]: [UI] [REPL] Interpolated 11/421 Frames (3%) - Average Speed: 3.30 FPS In / 6.60 FPS Out - Time: 02s - ETA: 01:02 [00001195] [04-04-2023 21:34:48]: [UI] [REPL] Interpolated 79/421 Frames (19%) - Average Speed: 18.13 FPS In / 36.26 FPS Out - Time: 03s - ETA: 09s [00001197] [04-04-2023 21:34:48]: Disk space check for 'E:/': 67.9 GB free. [00001198] [04-04-2023 21:34:49]: [UI] [REPL] Interpolated 147/421 Frames (35%) - Average Speed: 27.32 FPS In / 54.65 FPS Out - Time: 03s - ETA: 05s [00001200] [04-04-2023 21:34:49]: [UI] [REPL] Interpolated 215/421 Frames (51%) - Average Speed: 33.55 FPS In / 67.10 FPS Out - Time: 04s - ETA: 03s [00001202] [04-04-2023 21:34:50]: [UI] [REPL] Interpolated 281/421 Frames (67%) - Average Speed: 37.82 FPS In / 75.64 FPS Out - Time: 04s - ETA: 02s [00001226] [04-04-2023 21:34:50]: [UI] Done running RIFE - Interpolation took 05s. Peak Output FPS: 75.64 [00001227] [04-04-2023 21:34:50]: Muxing audio/subtitles into video... [00001228] [04-04-2023 21:34:50]: Skipping muxing from input step as there is no input video, only frames. [00001229] [04-04-2023 21:34:50]: [UI] Deleting temporary files...
[00001231] [04-04-2023 21:34:50]: [UI] Total processing time: 07s [00001232] [04-04-2023 21:34:50]: Done interpolating!
tinkertims
Hi, all. (crossposted from discussions, perhaps this was the better place to post? Sorry for the confusion)
A while back, I had some great success after being suggested I use Flowframes for a personal project of mine involving a famous sequence of photos taken by Gary Rosenquist, of the Mount St. Helens eruption in 1980. This project involved the use of AI-enhanced documentary screenshots of that sequence. (A link to that is here: https://www.youtube.com/watch?v=UNlP9TGZOMI )
The quality of those AI-enhanced screenshots, while great for the purpose of a basic illustration, did not match the original resolution of the photos Rosenquist took. At the time, I did not have them, and that was two months ago.
Since then, through several sources (mostly old, archived USGS reports and a few books), I've been able to source the original resolution copies of Rosenquist's famous sequence, and as a result, I'm attempting a full-HD scale interpolation in the way I did the last one.
When I attempt this process, the frames are subsequently mis-ordered and interpolated out of sequence. Attached is a screenshot of the raw source folder from which I am working (disregard the last frame, it's a hypothetical frame based on a photo by another observer named Keith Ronnholm, who was at the same location, taken after Rosenquist ran out of film). Each frame is ordered 1-22.
Subsequently, I've attached a screenshot showing the output interpolation folder. Notice both the out-of-order interpolation at which the frames are created, and the garbled appearance of some of the interpolated frames. I did not experience this when performing the earlier interpolation and in fact the earlier interpolation went off without a hitch. Now that I am working with higher-resolution source material, it seems as though the interpolation is flawed. Is this because of the higher-resolution source material?
More importantly, how do I fix it?