Add video support
3ec0ed86
iejMac
marked this pull request as draft 3 years ago
Merge branch 'main' into vivit
0403d1df
data loading: correct shapes in training loop (crappy code)
1cd33e41
Merge branch 'vivit' of https://github.com/iejMac/open_clip into vivit
107292fe
update model progress
b9957506
rename file + create_model loads something
be04c066
update
0ad71681
update
f9dfd024
embeddings get to loss, time to implement video encoding
df1c6983
update, set num_samples
7643ce3d
more filling in
cb12acdc
slightly improved preprocessing
67a2d336
update
f0615f9f
update weird lag
7bd848e2
simpler dataloader same results
80f41a06
properly normalize frames
f5af600d
update no temporal
982ee887
filter no mp4 samples
3f4fde7c
update
7176d95a
adding projection removes weird const loss bug but training doesn't g…
3f64b62c
some updates
8d3cc489
save changes
42304280
update dataloader to use video2dataset
9413c5e9
update
5c65a52d
repeat is bad
f2fa5bd3
enable loading CLIP weights to spatial and text encoders
31251717
update
b093a4c4
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub