Skip to content

Tags: hmmhmmhm/imagen-pytorch

Tags

v0.0.17

Toggle v0.0.17's commit message
add t5-base as usable text encoder (huggingface makes this easy)

v0.0.16

Toggle v0.0.16's commit message
only cosine and linear noise schedules used, just limit it to these 2…

…, so the continuous versions are easier to build

v0.0.15

Toggle v0.0.15's commit message
allow for config driven creation of imagen off the bat, also research…

…er must always give list of image resolutions this time around

v0.0.14

Toggle v0.0.14's commit message
allow for config driven creation of imagen off the bat, also research…

…er must always give list of image resolutions this time around

v0.0.12

Toggle v0.0.12's commit message
fix bug

v0.0.11

Toggle v0.0.11's commit message
add trainer class

v0.0.10

Toggle v0.0.10's commit message
complete the lowres augmentation level conditioning (which is basical…

…ly exactly like ddpm forward pass but for the lowres conditioning image). sampling is fixed at 0.2 (in paper they proposed 0.1 or 0.3)

v0.0.10a

Toggle v0.0.10a's commit message
allow for overriding the lowres conditioning noise level for sampling

v0.0.9

Toggle v0.0.9's commit message
complete the lowres augmentation level conditioning (which is basical…

…ly exactly like ddpm forward pass but for the lowres conditioning image). sampling is fixed at 0.2 (in paper they proposed 0.1 or 0.3)

v0.0.8

Toggle v0.0.8's commit message
more t5 fixes