SmilingWolf
commited on
Commit
•
805ca98
1
Parent(s):
9e6d6a7
Update model files
Browse files- README.md +15 -2
- model.msgpack +1 -1
- model.onnx +1 -1
- model.safetensors +1 -1
- sw_jax_cv_config.json +1 -1
README.md
CHANGED
@@ -17,11 +17,16 @@ Images with less than 10 general tags were filtered out.
|
|
17 |
Tags with less than 600 images were filtered out.
|
18 |
|
19 |
## Validation results
|
20 |
-
`P=R: threshold = 0.
|
|
|
21 |
|
22 |
## What's new
|
|
|
|
|
|
|
|
|
23 |
Model v1.1/Dataset v3:
|
24 |
-
Amended the JAX model config file: add image size.
|
25 |
No change to the trained weights.
|
26 |
|
27 |
Model v1.0/Dataset v3:
|
@@ -31,6 +36,14 @@ ONNX model is compatible with code developed for the v2 series of models.
|
|
31 |
The batch dimension of the ONNX model is not fixed to 1 anymore. Now you can go crazy with batch inference.
|
32 |
Switched to Macro-F1 to measure model performance since it gives me a better gauge of overall training progress.
|
33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
## Final words
|
35 |
Subject to change and updates.
|
36 |
Downstream users are encouraged to use tagged releases rather than relying on the head of the repo.
|
|
|
17 |
Tags with less than 600 images were filtered out.
|
18 |
|
19 |
## Validation results
|
20 |
+
`v2.0: P=R: threshold = 0.2653, F1 = 0.4541`
|
21 |
+
`v1.0: P=R: threshold = 0.2521, F1 = 0.4411`
|
22 |
|
23 |
## What's new
|
24 |
+
Model v2.0/Dataset v3:
|
25 |
+
Trained for a few more epochs.
|
26 |
+
Used tag frequency-based loss scaling to combat class imbalance.
|
27 |
+
|
28 |
Model v1.1/Dataset v3:
|
29 |
+
Amended the JAX model config file: add image size.
|
30 |
No change to the trained weights.
|
31 |
|
32 |
Model v1.0/Dataset v3:
|
|
|
36 |
The batch dimension of the ONNX model is not fixed to 1 anymore. Now you can go crazy with batch inference.
|
37 |
Switched to Macro-F1 to measure model performance since it gives me a better gauge of overall training progress.
|
38 |
|
39 |
+
# Runtime deps
|
40 |
+
ONNX model requires `onnxruntime >= 1.17.0`
|
41 |
+
|
42 |
+
# Inference code examples
|
43 |
+
For timm: https://github.com/neggles/wdv3-timm
|
44 |
+
For ONNX: https://huggingface.co/spaces/SmilingWolf/wd-tagger
|
45 |
+
For JAX: https://github.com/SmilingWolf/wdv3-jax
|
46 |
+
|
47 |
## Final words
|
48 |
Subject to change and updates.
|
49 |
Downstream users are encouraged to use tagged releases rather than relying on the head of the repo.
|
model.msgpack
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 413777297
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:a26ae4edbe45fa155ede34257153642b3bf476044a13e0ae6a0c4c131198f444
|
3 |
size 413777297
|
model.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 467460978
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e6774bff34d43bd49f75a47db4ef217dce701c9847b546523eb85ff6dbba1db1
|
3 |
size 467460978
|
model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 392149220
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:39fe04000d5907a0a7616df54af744d43fbd70b5fba7abc3135d3675734c03ef
|
3 |
size 392149220
|
sw_jax_cv_config.json
CHANGED
@@ -16,4 +16,4 @@
|
|
16 |
"patch_norm": true,
|
17 |
"layer_norm_eps": 1e-05
|
18 |
}
|
19 |
-
}
|
|
|
16 |
"patch_norm": true,
|
17 |
"layer_norm_eps": 1e-05
|
18 |
}
|
19 |
+
}
|