I like hina too
Social Media Girls
-
Going forward please follow these rules when posting:
-The content MUST be from TikTok, do not post slips from other platforms.
-You MUST include profile names / link to the profile, or make every attempt to include it.
-Include timestamps if you post a long video or a full live video.
-Make sure ALL girls posted are 18+
-
Please make sure all of the videos and screenshots you're posting are only of people over the age of 18.
This also means no children in the background.
No more warnings.
Ionic Spark commented
1m iterations
sourcing facesets is difficult, there are a lot of ways but here's mine:
1. I started selecting/downloading videos from her tiktok, twitter, youtube (@hapycb) and their QWER YT channel (with good closeups)
2. Compile and Cut the videos with closeups and good yaw distribution, then export. (if you have AI Video enhancer use it and skip No.4)
On DFL:
3. '2) extract images from video data_src' then '4) data_src faceset extract' (in here you will set the size depending on your pc's computing capabilities, for me I go with whole face/ 512x512 cause my pc is trash) - Go to /workspace/data_src/aligned folder and delete blurred/extra face images.
4. This is optional but this will increase better facial artifact in training:
- select '4.2) data_src util faceset metadata save' and locate 'meta.dat' inside aligned folder, make a new folder on this path /workspace/data_src name it enhanced_aligned and paste the meta.dat file inside.
- If you have Gigapixel 7 or 8 enhance the whole aligned folder and output with same input setting (512x512) - you could export the enhanced imgs to 'aligned_enhanced' folder
- once finished, delete ng aligned folder and rename "enhanced_aligned" folder to "aligned" then '4.2) data_src util faceset metadata restore'
now you have enhanced faceset!
but in the merging part you mention, I guess there are misaligned on labeled dst? would you like me to explain how to prepare data_dst?
There's also Training, Xseg (optional) and Merging.
sourcing facesets is difficult, there are a lot of ways but here's mine:
1. I started selecting/downloading videos from her tiktok, twitter, youtube (@hapycb) and their QWER YT channel (with good closeups)
2. Compile and Cut the videos with closeups and good yaw distribution, then export. (if you have AI Video enhancer use it and skip No.4)
On DFL:
3. '2) extract images from video data_src' then '4) data_src faceset extract' (in here you will set the size depending on your pc's computing capabilities, for me I go with whole face/ 512x512 cause my pc is trash) - Go to /workspace/data_src/aligned folder and delete blurred/extra face images.
4. This is optional but this will increase better facial artifact in training:
- select '4.2) data_src util faceset metadata save' and locate 'meta.dat' inside aligned folder, make a new folder on this path /workspace/data_src name it enhanced_aligned and paste the meta.dat file inside.
- If you have Gigapixel 7 or 8 enhance the whole aligned folder and output with same input setting (512x512) - you could export the enhanced imgs to 'aligned_enhanced' folder
- once finished, delete ng aligned folder and rename "enhanced_aligned" folder to "aligned" then '4.2) data_src util faceset metadata restore'
now you have enhanced faceset!
but in the merging part you mention, I guess there are misaligned on labeled dst? would you like me to explain how to prepare data_dst?
There's also Training, Xseg (optional) and Merging.
Ionic Spark commented
I don't usually take requests but the Jav is good and has a lot of closeup. unfortunately, my model cannot handle those and the face may become unrecognizable after merging :< (I could try tho)
plz I'll definitely ask you a favor bro