sahirp commited on
Commit
3c34472
·
verified ·
1 Parent(s): d8454ad

Upload folder using huggingface_hub

Browse files
Files changed (4) hide show
  1. README.md +70 -0
  2. best_model_v2.pt +3 -0
  3. product_embeddings.pkl +3 -0
  4. scene_embeddings.pkl +3 -0
README.md ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags:
4
+ - fashion
5
+ - recommendation
6
+ - pytorch
7
+ - image-to-image
8
+ datasets:
9
+ - shop-the-look
10
+ language:
11
+ - en
12
+ ---
13
+
14
+ # complete the look - fashion compatibility model
15
+
16
+ scene-based complementary product recommendation model. given an outfit photo, find products that match the style.
17
+
18
+ ## model description
19
+
20
+ trained on the [shop the look dataset](https://github.com/kang205/STL-Dataset) with fashion SigLIP embeddings.
21
+
22
+ ### architecture
23
+ - **input**: fashion SigLIP embeddings (768-dim)
24
+ - **scene encoder**: MLP with residual blocks → 256-dim
25
+ - **product encoder**: MLP with category embeddings → 256-dim
26
+ - **training**: contrastive + triplet loss with hard negative mining
27
+
28
+ ### performance
29
+
30
+ | metric | score |
31
+ |--------|-------|
32
+ | R@1 | 12.3% |
33
+ | R@5 | 28.6% |
34
+ | R@10 | 36.6% |
35
+ | MRR | 0.204 |
36
+
37
+ ## files
38
+
39
+ - `best_model_v2.pt` - model weights (37MB)
40
+ - `product_embeddings.pkl` - pre-computed product embeddings (110MB, 36K products)
41
+ - `scene_embeddings.pkl` - pre-computed scene embeddings (87MB, 29K scenes)
42
+
43
+ ## usage
44
+
45
+ ```python
46
+ import torch
47
+ import pickle
48
+
49
+ checkpoint = torch.load("best_model_v2.pt", map_location="cpu", weights_only=False)
50
+ # see github repo for full code
51
+
52
+ with open("product_embeddings.pkl", "rb") as f:
53
+ product_embeddings = pickle.load(f)
54
+ ```
55
+
56
+ ## code
57
+
58
+ full training code and web demo: [github repo](https://github.com/USER/complete-the-look)
59
+
60
+ ## citation
61
+
62
+ based on:
63
+ ```bibtex
64
+ @inproceedings{kang2019complete,
65
+ title={Complete the Look: Scene-based Complementary Product Recommendation},
66
+ author={Kang, Wang-Cheng and Kim, Eric and Leskovec, Jure and Rosenberg, Charles and McAuley, Julian},
67
+ booktitle={CVPR},
68
+ year={2019}
69
+ }
70
+ ```
best_model_v2.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f8ea650daf368ed2db9c0037b0eb06d9ac787780a969eb46e9fbdea76c062863
3
+ size 38396227
product_embeddings.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6bb72fe85a731311957c081099317583e51b655a572dba5176daa95713c96e7a
3
+ size 115086425
scene_embeddings.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe1c7e1dc5d2a0e26af27703d93d9edda82b4837c47e1c4fdc95b2bce60fbd08
3
+ size 90445039