SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2205.01580 · 2022

Better plain ViT baselines for ImageNet-1k

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 24 functions out of this paper's own repositories and ran 23 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
yuyangshu/retinavit pwc_unofficial 17 of 18
copy not recorded — 4 of 4
osiriszjq/impulse_init extension 1 of 1
conceptofmind/Simple-ViT-flax reimplementation 1 of 1
FunctionStatusWhere it lives
Model Ran yuyangshu/retinavit/big_vision/models/mlp_mixer.py
code served (permissive licence) · get_code("0a92f4de53a489b4")
SpatialConv2d_Linear_init Ran this paper's copy was not recorded; identical code first harvested from osiriszjq/impulse_init
pointer only · get_code("0440e95034d883bd")
SpatialConv2d_init Ran this paper's copy was not recorded; identical code first harvested from osiriszjq/impulse_init
pointer only · get_code("b04417d600247aaa")
dot_product_attention Ran yuyangshu/retinavit/big_vision/models/attn_override.py
code served (permissive licence) · get_code("2c4ed3f75ab6ee14")
dot_product_attention_weights Ran yuyangshu/retinavit/big_vision/models/attn_override.py
code served (permissive licence) · get_code("c8927d5d53a840b6")
fix_old_checkpoints Ran yuyangshu/retinavit/big_vision/models/bit.py
code served (permissive licence) · get_code("75f9a36fc8351778")
get Ran yuyangshu/retinavit/big_vision/datasets/core.py
code served (permissive licence) · get_code("f40ba1fc76a56621")
get_block_desc Ran yuyangshu/retinavit/big_vision/models/bit.py
code served (permissive licence) · get_code("4f697aeaa3747cc2")
get_count Ran yuyangshu/retinavit/big_vision/optax.py
code served (permissive licence) · get_code("5c44a598f86dadfc")
get_posemb Ran yuyangshu/retinavit/big_vision/models/vit.py
code served (permissive licence) · get_code("ec1bf067066bb434")
impulse_init Ran osiriszjq/impulse_init/vit.py
code served (permissive licence) · get_code("ce9f577538e842e0")
magnitude Ran yuyangshu/retinavit/big_vision/models/attn_override.py
code served (permissive licence) · get_code("9d78a212d314b83e")
make_for_inference Ran yuyangshu/retinavit/big_vision/input_pipeline.py
code served (permissive licence) · get_code("a0fec96972fe8326")
npload Ran yuyangshu/retinavit/big_vision/utils.py
code served (permissive licence) · get_code("2b3549774154825a")
onehot Ran yuyangshu/retinavit/big_vision/utils.py
code served (permissive licence) · get_code("7b17682564107576")
pad_shard_unpad Ran yuyangshu/retinavit/big_vision/utils.py
code served (permissive licence) · get_code("5ad1f39431340fbb")
pair Ran this paper's copy was not recorded; identical code first harvested from LINs-lab/DynMoE
pointer only · get_code("6ba8cee9f5daea41")
posemb_conv Ran yuyangshu/retinavit/big_vision/models/vit.py
code served (permissive licence) · get_code("4f5dd3aec9da6ede")
posemb_sincos_2d Ran this paper's copy was not recorded; identical code first harvested from lucidrains/vit-pytorch
pointer only · get_code("cfe18b9cc8afbfe9")
posemb_sincos_2d Ran conceptofmind/Simple-ViT-flax/simple_vit.py
code served (permissive licence) · get_code("b848f8a0a583e58b")
posemb_sincos_2d Ran yuyangshu/retinavit/big_vision/models/vit.py
code served (permissive licence) · get_code("2d67a8a756fe7cf8")
standardize Ran yuyangshu/retinavit/big_vision/models/bit_paper.py
code served (permissive licence) · get_code("c272f024d815d59d")
weight_standardize Ran yuyangshu/retinavit/big_vision/models/bit.py
code served (permissive licence) · get_code("37b141b05f87de19")
find_states Not yet run yuyangshu/retinavit/big_vision/optax.py
code served (permissive licence) · get_code("1f64c95d3b75c46b")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

It is commonly accepted that the Vision Transformer model requires sophisticated regularization techniques to excel at ImageNet-1k scale data. Surprisingly, we find this is not the case and standard data augmentation is sufficient. This note presents a few minor modifications to the original Vision Transformer (ViT) vanilla training setting that dramatically improve the performance of plain ViT models. Notably, 90 epochs of training surpass 76% top-1 accuracy in under seven hours on a TPUv3-8, similar to the classic ResNet50 baseline, and 300 epochs of training reach 80% in less than one day.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2205.01580")
get_code_for_paper("2205.01580")
have("2205.01580")

Connect an agent — have() is free.