LTH14/mage

Issues

• About problems in the training process

  #57 opened 3 months ago by 18222why
  1

• A issues about ckpt

  #56 opened 3 months ago by 18222why
  3

• some questions about changing the image classification data set

  #45 opened a year ago by cb-rep
  11

• Request for the implementation of "Autoregressive Image Generation without Vector Quantization"

  #54 opened 5 months ago by iamlockelightning
  3

• Regarding codes reconstructing masked images

  #55 opened 6 months ago by Haoyu1Chen
  0

• A question about vocab_size in token embbeding

  #53 opened 8 months ago by tanbuzheng
  23

• Is there any script to generate the Fig.2 in the paper?

  #15 opened 2 years ago by LinB203
  13

• Cannot load the fine-tuned ckpt

  #51 opened a year ago by hlzhang109
  4

• Why not adopt bert like MaskGIT to reconstruct Tokens?

  #52 opened a year ago by tanbuzheng
  4

• About test.py for testing the performance of the pretrain model

  #33 opened 2 years ago by AbandonedWarlord
  2

• 图像重建

  #50 opened a year ago by yaoqi-liu
  0

• Question on image Inpainting

  #49 opened a year ago by GGGGxxxxxxxxr
  6

• How to test the outpainting or inpainting of images

  #39 opened a year ago by 83517769
  3

• Question on Masking ratio

  #47 opened a year ago by sukun1045
  5

• Query About Figure 3 in the MAGE Article

  #48 opened a year ago by Yuan1z0825
  1

• Some question about change datasets

  #42 opened a year ago by zhihao-2022
  3

• Question about applying to my own dataset

  #46 opened a year ago by mabaisen
  2

• Are cls_token and pos_embed not used in models_mage.py?

  #38 opened a year ago by yxchng
  3

• Why did you retrain a new module token_emb, have you ever tried to change the VQGAN codebook from dim=256 to dim=768 via MLP?

  #24 opened 2 years ago by BinZhu-ece
  2

• The specified path cannot be found

  #41 opened a year ago by zhanglaoban-kk
  10

• Difference between mage and maskgit?

  #44 opened a year ago by lqniunjunlper
  0

• How sensitive is this model to different batch size?

  #35 opened a year ago by yxchng
  2

• reconstructed images

  #28 opened 2 years ago by luodi789567890
  3

• 'vqgan_jax_strongaug.ckpt'，Can you provide a model of VQgan for pre training.

  #43 opened a year ago by Liuhp133
  0

• Could you provide a re-implement with one GPU?

  #30 opened 2 years ago by Beyondzjl
  2

• Error While Running the Pre-Train Command on a local Machine

  #40 opened a year ago by SaadH-077
  1

• Is it normal for inception score to be different?

  #36 opened a year ago by yxchng
  1

• Do you mind providing the conversion script for jax vqgan checkpoints?

  #37 opened a year ago by yxchng
  0

• About releasing the pretrained weights trained with weak augmentation

  #32 opened 2 years ago by junwuzhang19
  1

• About Post-training Evaluation

  #34 opened a year ago by Neyleer
  2

• JAX version code

  #10 opened a year ago by magicknight
  1

• Can't find the file

  #11 opened a year ago by zhangjinglin23
  1

• question about Class-Conditional Image Generation

  #16 opened 2 years ago by LinB203
  3

• How to get the unconditional generation result of Maskgit in Table 1?

  #25 opened 2 years ago by junwuzhang19
  1

• Why encoder-decoder architecture?

  #31 opened 2 years ago by bsxxdw
  1

• AssertionError happens when loading the model in the workflow of “finetuning-then-linprobing”.

  #29 opened 2 years ago by uk9921
  3

• error

  #27 opened 2 years ago by luodi789567890
  4

• Forcing FP32 in Attention

  #26 opened 2 years ago by gshaikov
  1

• About your VQGAN codebase and model

  #22 opened 2 years ago by junwuzhang19
  1

• The data domain is different.

  #23 opened 2 years ago by LinB203
  1

• Mismaching values for VQGAN state_dict

  #21 opened 2 years ago by ImmortalSdm
  6

• About the computation of logits after decoder

  #20 opened 2 years ago by liuqk3
  1

• Could you provide a training log?

  #18 opened 2 years ago by LinB203
  1

• unconditional class generation

  #19 opened 2 years ago by zhangjinglin23
  1

• Reconstruction looks terrible.

  #17 opened 2 years ago by ArneNx
  6

• How to determine the value of fake_class_label in models_mage.py

  #12 opened 2 years ago by xcyuyuyu
  3

• Is the decoder code redundant?

  #14 opened 2 years ago by LinB203
  4

• Why does the encoder need a little bit of mask token?

  #13 opened 2 years ago by LinB203
  4

• Some questions about the VQGAN tokenizer

  #9 opened 2 years ago by LooperXX
  10

• slightly different performance result of un-conditional generation on ImageNet

  #8 opened 2 years ago by ChangyaoTian
  2