1adrianb / 1adrianb/face-alignment

Request: Shouldn't require much assistance (almost zero assistance), with reference to using the full shape, expression model.

Offen
#363 1 Kommentar 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen
Vorherrschende Sprache
Python
Sterne
7.5k
Forks
1.4k
PR-Merge-Kennzahlen
Keine gemergten PRs in 30 T.

Beschreibung

Hello, excellent paper excellent repository.

I have seen a few requests for the full model, rather than 62 parameters (12 pose, 40 shape, 10 expression).
Can you provide the dull 3D morphable model (12 pose, 199 shape and 29 expression, or is it available via the BFM model site)?

If you can point me at (or provide) the full model I might be able to provide additional training code to fit the full model (or a user defined subset of it).

I ask, because the paper shows some excellent inventive steps: use of projected cylindrical convolution & parameter fitting order. I believe that this could produce a significant increase in accuracy for face reconstruction; both shape and texture.

I do not require much help (probably zero, but one never knows), just access to the data required to train the full model parameter fitting.

Can you provide any assistance?

Beitragsleitfaden

Für dieses Repository ist kein Beitragsleitfaden indexiert

Rechercherichtung

The issue requests access to a full 3D morphable model (199 shape, 29 expression parameters) for training. Check the repository's model loading code and data sources, likely referencing the Basel Face Model (BFM). Look for existing model files or scripts in the repo, and see if the maintainer has provided any guidance on obtaining extended models. The goal is to enable training with more parameters, which involves understanding the model structure and fitting pipeline.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Bewertung

Tech-Stack
python, pytorch
Bereich
ai, computer-vision, machine-learning
Issue-Typ
Feature
Schwierigkeit
5/5
Geschätzter Aufwand
Über eine Woche
Aktivitätsstatus
Veraltet
Klarheit
Größtenteils klar
Anfängerfreundlichkeit
25/100

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.