ajehsmihba commited on
Commit
e8b6209
·
verified ·
1 Parent(s): 5298d66

Upload folder using huggingface_hub

Browse files
Files changed (2) hide show
  1. best_checkpoint.pth +3 -0
  2. readMe.md +70 -0
best_checkpoint.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:826a00120a0510635d90be877ed76fa23270bf19bb81fec6dcc6294c65e31288
3
+ size 403768367
readMe.md ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - MuseMachine
4
+ - pytorch
5
+ - generative-adversarial-network
6
+ - variational-autoencoder
7
+ ---
8
+
9
+ # MuseMachine
10
+
11
+ A modular AE-GAN pipeline for portrait generation.
12
+
13
+ ## Architecture
14
+
15
+ | Component | Description |
16
+ |---|---|
17
+ | Encoder | VAE encoder producing latent representations |
18
+ | Generator | Decodes latent vectors into coarse images |
19
+ | Refiner | Iteratively refines generations over 4 steps |
20
+ | Discriminator | PatchGAN-style discriminator |
21
+
22
+ ## Hyperparameters
23
+
24
+ ```json
25
+ {
26
+ "image_size": 256,
27
+ "latent_dim": 256,
28
+ "batch_size": 4,
29
+ "epochs": 100,
30
+ "lr_g": 0.0002,
31
+ "lr_d": 5e-05,
32
+ "alpha_recon": 100.0,
33
+ "beta_adv": 1.0,
34
+ "beta_kl": 0.05,
35
+ "noise_std": 0.15,
36
+ "refinement_steps": 4,
37
+ "refiner_step_size": 0.08,
38
+ "refine_decay": 0.8,
39
+ "dataset_name": "ajehsmihba/aesthetic-female-portraits",
40
+ "outputs_dir": "outputs",
41
+ "models_dir": "models",
42
+ "save_every": 50,
43
+ "push_to_hf": true,
44
+ "hf_repo_id": "ajehsmihba/MuseMachine",
45
+ "grad_clip": 1.0,
46
+ "vis_every": 10
47
+ }
48
+ ```
49
+
50
+ ## Best Metrics
51
+
52
+ ```json
53
+ {
54
+ "D": 1.0980566523291848,
55
+ "EG": 88.96168032559481,
56
+ "KL": 337.3994445800781,
57
+ "Recon": 0.1864572438326749,
58
+ "Coarse_Recon": 0.19136723740534348
59
+ }
60
+ ```
61
+
62
+ ## Resume Training
63
+
64
+ ```python
65
+ from huggingface_hub import hf_hub_download
66
+ import torch
67
+
68
+ ckpt_path = hf_hub_download(repo_id="ajehsmihba/MuseMachine", filename="best_checkpoint.pth")
69
+ ckpt = torch.load(ckpt_path, map_location="cpu")
70
+ ```