yangnaruto / latent-energy-transport Goto Github PK

View Code? Open in Web Editor NEW

47.0 47.0 9.0 86 MB

Python 99.78% Shell 0.22%

latent-energy-transport's People

Stargazers

Watchers

Forkers

josephkj dlwbm123 gnehz comzzw mtjmtj7 hello-trouble pphuangyi jxzhangjhu terry97-guel

latent-energy-transport's Issues

Proof-of-Concept experiment

Hi Yang,

Thanks for this amazing work, and for sharing the code.

Have you included the Proof-of-Concept experiment (Sec 4.2 in the paper) code in this repo? It is such a clean experiment, it would be very nice to have that!

Thanks in advance,
Joseph

Hi, thanks for your wonderful job!

I have a little question. What is the files under the datasetpath? I run the prepare_celeba_hq_tfrecords.py, and it generates many tfrecords files. I don't know how to set the datasetpath here. Could you help me? Thank you very much!

RuntimeError: main thread is not in main loop

Hi, I'm trying to train ALAE and I was able to train for 2 epochs, then

2022-01-19 12:19:33,395 logger INFO: Saving checkpoint to training_artifacts/solidAHdot/model_tmp_lod0.pth
2022-01-19 12:19:33,395 logger INFO: 
[3/500] - ptime: 12.91, loss_d: 1.5513756, loss_g: 0.8106965, lae: 0.3998570, blend: 1.000, lr: 0.001500000000,  0.001500000000, max mem: 832.148438",
2022-01-19 12:19:33,399 logger INFO: Batch size: 128, Batch size per GPU: 128, LOD: 0 - 4x4, blend: 1.000, dataset size: 16480
/home/qimin/Projects/latent-energy-transport/ALAE/tracker.py:109: UserWarning: Starting a Matplotlib GUI outside of the main thread will likely fail.
  plt.figure(figsize=(12, 8))
Saved to training_artifacts/solidAHdot/sample_4_0.jpg
2022-01-19 12:19:46,812 logger INFO: Saving checkpoint to training_artifacts/solidAHdot/model_tmp_lod0.pth
2022-01-19 12:19:46,813 logger INFO: 
[4/500] - ptime: 13.41, loss_d: 1.5081978, loss_g: 0.7740349, lae: 0.4387354, blend: 1.000, lr: 0.001500000000,  0.001500000000, max mem: 832.148438",
2022-01-19 12:19:46,817 logger INFO: Batch size: 128, Batch size per GPU: 128, LOD: 0 - 4x4, blend: 1.000, dataset size: 16480
Saved to training_artifacts/solidAHdot/sample_5_0.jpg
Exception ignored in: <bound method Image.__del__ of <tkinter.PhotoImage object at 0x7fe2a86befd0>>
Traceback (most recent call last):
  File "/home/qimin/anaconda3/envs/letit/lib/python3.6/tkinter/__init__.py", line 3507, in __del__
    self.tk.call('image', 'delete', self.name)
RuntimeError: main thread is not in main loop

Did this ever happen to you before? Thanks

note: I used my own dataset which is binary image (grayscale), not sure if this is the issue but I was able to run prepare_data.py

A kind request for clarification

Hi Yang Zhao (@YangNaruto),

Thanks again for your amazing work and elegant code base. It would be very kind of you if you could help me clarify one aspect of the loss formulation.

The equation 2 reads as follows:

This nicely translates to:

loss = -(target_energy - source_energy).mean() in code: [1], [2].

The second part of equation 2 is an expectation over samples from the EBM, which are sampled using Langevin dynamics as follows:

z_src_q = langvin_sampler(ebm, z_src)
source_energy = ebm(z_src_q)

Why do we start the Langevin sampling from the source domain's z (z_src)? I will explain what indeed confuses me. Equation 2 basically tries to learn the energy manifold of the target domain. Why do we use the source domain information to learn the target energy distribution?

yangnaruto / latent-energy-transport Goto Github PK

latent-energy-transport's People

Stargazers

Watchers

Forkers

latent-energy-transport's Issues

Proof-of-Concept experiment

TFRecordsDataset

RuntimeError: main thread is not in main loop

A kind request for clarification

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent