facebookresearch / mae

PyTorch implementation of MAE https//arxiv.org/abs/2111.06377

facebookresearch/mae Issues

Two different checkpoints for each ViT type
Closed a month ago5
Is the training procedure result normal? Masked regions do not improve and appear to be random noise.
Updated a month ago1
Reconstruction using normalized pixel values to get unnormalized pixel values?
Updated a month ago
Loss is considerably worse on custom data set with different mean and standard deviation
Updated a month ago2
训练的代码用最新的timm跑不通
Closed a month ago
About the gan-loss
Updated 2 months ago2
Can run interactive visualization demo with GPU?
Updated 2 months ago
collab notebook error
Updated 2 months ago2
How to obtain the complete reconstructed image?
Updated 2 months ago
Poor Image Reconstruction After Fine-tuning on MvTec Dataset
Closed 2 months ago2
Could you provide the pretrained checkpoints of both encoder and decoder in MAE?
Closed 2 months ago2
Code: Compatible to any channels for function patchify and unpatchify
Closed 3 months ago
model.fc_norm is not trained in linear probing
Closed 4 months ago
Not able to import inf from torch._six
Closed 10 months ago4
Is possible to enable FP16 or TF32 in pretraining?
Updated 4 months ago1
visualization attention map.
Updated 4 months ago
Small naming error - masking generation
Updated 5 months ago1
param_groups_lrd for layer decay
Updated 5 months ago1
Does the MAE pre-trained model transfer well to CIFAR？
Updated 5 months ago1
Question about `len_keep` at random masking function
Updated 5 months ago1
A question about DropPath in pretraining
Updated 5 months ago1
I found both LLAMA and MAE used smaller beta2 in ADAMW optimizer during pre-training. Is that any intuition behind such setting?
Updated 5 months ago1
Error in loading pretrained weight for 'mae_vit_base_patch16'
Updated 5 months ago2
How to obtain the reconstructed image for inference and masked
Updated 6 months ago
patchify and unpatchify
Updated 6 months ago1
Is the visualization result normal?
Updated 6 months ago2
pretrain error for import timm
Closed a year ago2
Question about PatchEmbed's Initialization Trick
Updated 8 months ago1
Confusion in The Loss Function Implementation.
Updated 10 months ago
Ask for segmentation finetune code
Updated 10 months ago
Creative Commons does not recommend their licenses be used for software
Updated 10 months ago
How to reconstruct some unlabeled images
Updated 10 months ago1
Dataset for "Fine-Tuning Pre-trained MAE for classification"
Updated 10 months ago
[Question] Reconstructed pixels are discontinuous at patch boundaries.
Updated a year ago9
The reconstruction of mae_pretrain_vit_base.pth is awful, is it right?
Closed a year ago1
Monitor training of custom dataset
Updated a year ago
AssertionError: Input image height (736) doesn't match model (224).
Updated a year ago1
Training time
Updated a year ago1
Question regarding figure 5 on the paper
Updated a year ago
License questions
Updated a year ago1
Single machine multi-GPU training
Updated a year ago2
Release of MAE decoder
Closed a year ago2
Implementing VIT Small in MAE
Updated a year ago1
Non-square number of patches
Updated a year ago
Shouldn't the patch embeddings be trained only on the patches that survived masking? (Rather than the original image)
Closed a year ago5
How about fine-tuning with MAE auxiliary task?
Updated a year ago
[Question] Ablation of encoder with mask token
Updated a year ago
[Question] Why non-masked patches look worse in pixel reconstruction example image
Updated a year ago1
does small batch size influence the training?
Updated a year ago
Something about the training set
Updated a year ago