First off, congrats on the new manual tool - I LOVE it!! It's a huge improvement over the old one - and the old one wasn't bad, once you got used to it. But the new one is SO much better - and faster to make adjustments - well done chaps!!
However, since the updates in the last few hours, it now constantly crashes after every ten faces - or less. And if I haven't saved it after every few, then I lose everything I've done - frustrating. Also, when I click the face to add a new alignment, the bounding box doesn't appear until I re-adjust the size of the screen by moving the divider between the upper panel and the lower panel - which slows things down even more.
I'm using the vgg_obstructed option when extracting - could there be an issue with this...? I haven't tried it with any other opton as yet, as up til now, this was just the easiest option to roll with overall...
I've also noticed that the majority of the alignments APPEAR to be correct in the lower panel when first opening the manual tool - but when I press "play", all the lower panels go black, indicating that the alignments are in the frame somewhere, but not actually on the faces at all...!
I've found that by repositioning these (which I can only do by making the upper panel really small, and seeing that the alignment is way off to the side somewhere...! and grabbing it and dragging it over to the correct spot) or, by deleting ALL the "faces"/ alignments within the tool, and then re-adding them one by one, I can get around this problem. But, as you can imagine, this is not ideal and VERY time consuming - even more time consuming than the old manual tool was...!!
I fully appreciate that this is not an easy task to program/ code this, and that there will always be hitches, problems and bugs to iron out, and that this is constantly in motion and constantly being upgraded by your goodselves, so I am VERY grateful for all the work that you guys do!
But I just wondered if this was a known problem, or just a recent thing?
Another thing that I wondered was that I currently have 2x 2070 supers running (no NVLink - yet...it's on it's way...) - could this affect it in any way? Or maybe the version of Python (3.8.4) or Anaconda (3.7.4) that I have running could be the issue and needs to be updated/ downgraded?
Many thanks for any light that you could shed on this!
Sometimes the thumbnails can go screwy, which can manifest itself in the way you describe, so try the regen-thumbs option when launching, and see if it fixes.
If it doesn't, a screen recording of the issue would be really useful. If you're on Windows, then you can capture this with Windows+G key combination.
My word is final
The only other thing I missed in my original post, is that the last time I didn't fix these, one face at a time, as stated, and chose to ignore that and see what it did, it worked fine! But this time, I tried to ignore it again and train it as it was - and it was stating that there were no faces in any of the frames...! Which is why I thought I'd have to go the frame-by-frame route again...
I will come back to you if the problem doesn't resolve itself after trying the regen-thumbs as you suggested...
Thanks for your time!
I have an issue as well with the alignment tool. It is a major improvement overall, and I think you guys did an amazing job. I am now trying the regular components or extended faces, as I was getting issues using the vgg_obstructed, but I seem to be having a similar problem. I go to find faces that were not found in the extraction, and I select them as usual and save the alignments file. Then I do an extraction based on the new alignments file, and typically some of the faces I selected manually and extracted, are now "not in the alignments file". For example, I have 10 frames with faces, but its missing faces on frames 6, 7,8. I go find these using the manual tool and save the alignments file in the manual tool. Then when I re extract the ten faces, faces 6,7, and 8 are extracted, but FS doesn't recognize them in the alignments file(even though I just extracted them from that same file) and when I go to train or convert, It will tell me faces 6, 7, and 8 do not exist in the alignments file. This has been true no matter the mask used in the original extraction. To make it more confusing, some of the faces I manually "found" do extract and appear in the alignments file. I might be doing something stupid here, but it has been a bug I have kept encountering.
This one is weird, and not one that I can replicate...
So a couple of things, (just to rule out the obvious)
1) You are saving the alignments file when you have added new faces (save icon)?
2) The alignments file that you are reloading into the Manual Tool. This is definitely your latest alignments file and not a backup?
If it isn't any of the above issues, if you could describe exact, repeatable steps to recreate your issue, it would be hugely appreciated.
My word is final