本帖最后由 werran 于 2022-10-3 20:09 编辑
ae dims可能会影响清晰度，但不是决定清晰度的，决定清晰度的当然是分辨率。看见我红字标出的地方了么，影响的是模型整体学习能力。
另外你说512的模型用256的ae dims表情都对不上？你怕是根本没用过吧。
Autoencoder & Dimensions (Dims)An autoencoder is a neural network architecture capable of discovering structure within [unlabeled] data in order to develop a compressed representation of the input. Autoencoders are an unsupervised learning technique in which we leverage neural networks for the task of representation learning. -Jeremy Jordan, [color=var(--linkInitialColor)]Introduction to Autoencoders
Autoencoder, decoder, and encoder value control the model’s neural network dimensions, which directly affect the model’s ability to learn faces. [td]| DIM | DESCRIPTION | | Auto Encoder Dims | Auto encoder dimensions. Affects the overall ability of the model to learn faces. | | Inter Dims | Inter dimensions (AMP). Affects the overall ability of the model to learn faces. Set equal to or higher than Auto Encoder dims. | | Encoder Dims | Encoder dimensions. Affects the ability of the encoder (input) to ingest faces. | | Decoder Dims | Decoder dimensions. Affects the ability of the decoder (output) to recreate faces. | | Decoder Mask Dims | Decoder mask dimensions. Affects the quality of learned masks; may affect training. |
DeepFaceLab 2.0 Autoencoder / Encoder / Decoder Dimension Descriptions
| 
|Archiver|手机版|deepfacelab中文网
|网站地图
发表于 2022-10-3 19:10:43
