e1277af2ba
[bug] Fix interpolation of positional embeddings ( #378 )
...
Use size instead of scale factor to specify the output size of nn.interpolate(): this avoids any rounding issue leading to mismatching output size and consistently generate the same output size as with the previous kludge (from facebookresearch/dino#8 ).
2024-02-22 19:10:54 +01:00
2302b6bf46
Update vision_transformer.py ( #331 )
...
Account for register tokens in get_intermediate_layers
2023-12-01 18:12:17 +01:00
da4b3825f0
Lint
2023-10-27 07:33:32 -07:00
ad5a262b22
Update param_groups.py ( #283 )
...
* Update param_groups.py
Update lr decay rates for reg tokens
* Update param_groups.py
2023-10-27 16:30:19 +02:00
e203621e57
More top-level README updates
2023-10-27 15:31:51 +02:00
89272b5c0c
Add more prominent link to paper
2023-10-27 15:26:57 +02:00
9c7e324579
Add new backbones trained with registers ( #282 )
...
Add new backbones (and matching linear classification heads) trained with 4 registers following [Vision Transformers Need Registers](https://arxiv.org/abs/2309.16588 ).
2023-10-27 15:15:10 +02:00
44abdbe27c
Fix interpolate parameters to allow tracing ( #247 )
...
Pass scale factor as a tuple of floats to F.interpolate() to allow tracing.
2023-09-30 22:29:41 +02:00
e7df9fc95d
Expose DPT depth models via torch.hub.load() ( #238 )
...
Add streamlined model versions w/o the mmcv dependency to directly load them via torch.hub.load().
2023-09-30 20:12:02 +02:00
82185b17a8
Expose linear depth models via PyTorch Hub ( #237 )
...
Add streamlined model versions w/o the mmcv dependency to directly load them via torch.hub.load().
2023-09-30 20:06:05 +02:00
b507fbcf50
Minor config tweaks ( #246 )
...
Ignore import warnings in hubconf and trigger lint workflow on PR.
2023-09-30 20:01:59 +02:00
9a4564ce5e
Rework Pytorch Hub support code ( #202 )
...
Rework support code for torch.hub.load() to allow reusing shared functions and eventually expose more models.
2023-09-27 17:06:03 +02:00
6a6261546c
Update README ( #189 )
...
Update the top-level README to make it clearer what's currently available.
2023-08-31 19:00:59 +02:00
dc1d2cbcc8
Fix broken links in notebooks
2023-08-31 09:43:55 -07:00
91d8cd81c2
Add semantic segmentation (Mask2Former) code ( #186 )
...
Add semantic segmentation (Mask2Former based on ViT-Adapter) code + update demo notebook for segmentation with a dedicated section.
2023-08-31 15:36:47 +02:00
d5b0405eff
Add semantic segmentation (linear) code ( #185 )
...
Add semantic segmentation (linear) code + demo notebook
2023-08-31 15:09:49 +02:00
d5c376b5b3
Add depth estimation code ( #184 )
...
Add depth estimation code + demo notebook
2023-08-31 14:57:50 +02:00
3a7bf1ca4b
Add (optional) extras dependencies ( #183 )
...
Add (optional) extras dependencies for dense tasks (mmcv and mmsegmentation) to conda and pip requirements.
2023-08-31 14:53:28 +02:00
81b2b64193
Update license everywhere ( #182 )
...
Update code and models license from CC-BY-NC to Apache 2.0 in headers and other files.
2023-08-31 14:41:52 +02:00
bd0bd9be19
Set default number of nodes to 1
...
We put default number of nodes to 1 in particular so that the linear evaluation with 8 GPUs has a global batch size of 1024, to reproduce the results.
2023-08-30 17:52:02 +02:00
ebc1cba109
Allow disabling xFormers via environment variable ( #180 )
...
Allow disabling the use of xFormers (for inference) by simply setting the XFORMERS_DISABLED environment variable
2023-08-30 17:20:47 +02:00
be7e57252f
Add missing hubconf arg ( #178 )
...
Add missing explicit layers argument
2023-08-29 16:05:32 +02:00
10d420147b
Remove mutable default arguments ( #170 )
...
Passing a list as default argument is not recommended.
2023-08-24 02:15:22 +02:00
84afc6fcce
Exclude venv from flake8 linting ( #168 )
...
Issue: flake8 processes the venv dir.
Change: added `exclude = venv` to flake8 conf
2023-08-24 02:14:23 +02:00
43c80c1ba8
Try to fix embedded video in README.md ( #165 )
2023-08-23 22:16:22 +02:00
c3c2683a13
Correct some typos ( #62 )
...
Signed-off-by: Alexander Seiler <seileralex@gmail.com>
2023-04-27 11:13:12 +02:00
c0ffb6ed71
Do not force using FlashAttention ( #58 )
...
Remove hardcoded selection of operator implementation and use xFormers fMHA dispatcher instead.
2023-04-26 02:26:24 +02:00
ca58ffcd87
Fix linear classifier wrapper ( #61 )
...
Fix linear classifier wrapper in PyTorch Hub configuration module to support multi-sample batches.
2023-04-26 02:15:45 +02:00
3e7e278d6f
Improve and fix ImageNet-1k dataset preparation ( #60 )
...
Document and fix implementation of extra metadata generation for ImageNet-1k.
2023-04-26 01:08:35 +02:00
4f26fee911
Datasets interface cleanup ( #59 )
...
Remove unused sample decoding interface in datasets.
2023-04-26 00:06:05 +02:00
f8969297db
Clarify some of the README.md contents ( #31 )
...
Installation, dependencies and Python module search path as well as a few other things.
2023-04-20 00:31:53 +02:00
719b1905cf
Minor tweaks to README.md ( #5 )
2023-04-19 23:02:15 +02:00
ea5276ec8c
Fix name in README ( #10 )
...
Fix author name.
2023-04-18 15:56:52 +02:00
fc49f49d73
Initial commit
2023-04-17 17:55:51 +00:00
Patrick Labatut
qasfb
Patrick Labatut
Leonid Ganeline
Alexander Seiler
Aryan Utkarsh