TencentARC/Open-MAGVIT2

Issues

• gpu训练很慢

  #39 opened 2 months ago by syyxsxx
  3

• 数据集格式怎么放

  #40 opened a month ago by Amber530
  1

• questions about the autoregressive model vocab size

  #41 opened 2 months ago by sysuyy
  2

• Noisy points in reconstruction results

  #12 opened 6 months ago by shinshiner
  4

• How long does it take to train this model?

  #7 opened 7 months ago by StarCycle
  3

• There is a problem with the inference result

  #38 opened 3 months ago by wen020
  1

• Could you improve the mushroom shape in the sample?

  #36 opened 3 months ago by TTPlanetPig
  1

• Could we make it a package?

  #33 opened 3 months ago by vinyesm
  3

• Any plan for 3D magvit-v2?

  #37 opened 4 months ago by txytju
  1

• Training loss curves?

  #18 opened 4 months ago by shinshiner
  5

• About the results without cfg

  #32 opened 4 months ago by youngsheen
  1

• Curious About Subspace Size

  #35 opened 4 months ago by Lqf-HFNJU
  2

• Better to put `self.model_ema = LitEma(self)` before `self.init_from_ckpt(...)`?

  #27 opened 4 months ago by AIWanderer-X
  1

• Can we write `_sigmoid_cross_entropy_with_logits` in a simple way?

  #21 opened 4 months ago by AIWanderer-X
  1

• Face loss for face reconstruction

  #31 opened 5 months ago by kl2004
  1

• UNet Model

  #30 opened 4 months ago by JeremyCJM
  1

• training bash script bug for 128 x 128

  #24 opened 4 months ago by deuterium1729
  0

• Which version of ImageNet?

  #29 opened 5 months ago by JeremyCJM
  2

• Questions about the update timeline

  #28 opened 5 months ago by Liam-lyr
  1

• Why not use `Adaptive GroupNorm` On Decoder & Sampler asymmetrical?

  #26 opened 5 months ago by xesdiny
  6

• `logits_real = logits_fake.reshape(B, -1)`: is this on purpose or just a bug?

  #20 opened 6 months ago by AIWanderer-X
  1

• Why not use `per_sample_entropy` loss for backward?

  #25 opened 6 months ago by xesdiny
  1

• Auto-Regressive Generative Model with MagVitV2 Tokenizer

  #19 opened 6 months ago by youngsheen
  1

• Could you Vs VQGAN-LC benchmark?

  #14 opened 6 months ago by xesdiny
  1

• No Checkpoint found after running run_128_B.sh

  #23 opened 6 months ago by wd255
  0

• Training phase sanity check fails by loading "../../data/imagenet/val" as an image

  #22 opened 6 months ago by wd255
  0

• Training phase sanity check fails by loading "../../data/imagenet/val" as an image

  #17 opened 6 months ago by wd255
  1

• Initializing distributed halts for a long time and then timeout

  #15 opened 6 months ago by wd255
  1

• Open-MAGVIT2 achieves SOTA??? I just cannot reproduce the numbers reported in the main page.

  #13 opened 6 months ago by juncongmoo
  16

• Will consider using Web images train a more robust version?

  #2 opened 6 months ago by MonolithFoundation
  11

• about the result of magvit2

  #11 opened 6 months ago by yyyouy
  1

• [BUG] reconstruct from index number instead of quant

  #10 opened 6 months ago by eisneim
  1

• ckpt ffhq

  #9 opened 6 months ago by kepengxu
  1

• Reconstruction result

  #1 opened 7 months ago by Epiphqny
  2

• Is it possible to also align magvit2 with a text encoder, as what you do in seed?

  #8 opened 7 months ago by StarCycle
  5

• [question] any plans to train higher compression ratio?

  #5 opened 7 months ago by eisneim
  3

• encoder和decoder的结果不一致。

  #6 opened 7 months ago by Mr-Harry
  3

• Inquiry Regarding Codebook Selection Method in Open-MAGVIT2

  #4 opened 7 months ago by zhaohm14
  2

• Do you compare your ckpt with stability's?

  #3 opened 7 months ago by StarCycle
  4