SwinV2 Tagger v2.1

README.md

@@ -16,8 +16,23 @@ Images with less than 10 general tags were filtered out.
 Tags with less than 600 images were filtered out.
 
 ## Validation results
-`P=R: threshold = 0.3771, F1 = 0.6854`
+`v2.0: P=R: threshold = 0.3771, F1 = 0.6854`
+
+## What's new
+Model v2.1/Dataset v2:  
+Re-exported to work around an ONNXRuntime v1.17.1 bug.  
+Bumped the minimum ONNXRuntime version to `>= 1.17.0`.  
+Now `timm` compatible! Load it up and give it a spin using the canonical one-liner!  
+Exported to `msgpack` for compatibility with the [JAX-CV](https://github.com/SmilingWolf/JAX-CV) codebase.  
+The batch dimension of the ONNX model is not fixed to 1 anymore. Now you can go crazy with batch inference.  
+No change to the trained weights themselves. There might be small prediction discrepancies across frameworks due to implementation details.  
+
+Model v2.0/Dataset v2:  
+Initial release.  
+
+# Runtime deps
+ONNX model requires `onnxruntime >= 1.17.0`
 
 ## Final words
 Subject to change and updates.  
-Downstream users are encouraged to use tagged releases rather than relying on the head of the repo.
+Downstream users are encouraged to use tagged releases rather than relying on the head of the repo.

config.json

@@ -0,0 +1,37 @@
+{
+  "architecture": "swinv2_base_window8_256",
+  "num_classes": 9083,
+  "num_features": 1024,
+  "global_pool": "avg",
+  "model_args": {
+    "act_layer": "gelu",
+    "img_size": 448,
+    "window_size": 14
+  },
+  "pretrained_cfg": {
+    "custom_load": false,
+    "input_size": [
+      3,
+      448,
+      448
+    ],
+    "fixed_input_size": false,
+    "interpolation": "bicubic",
+    "crop_pct": 1.0,
+    "crop_mode": "center",
+    "mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "num_classes": 9083,
+    "pool_size": null,
+    "first_conv": null,
+    "classifier": null
+  }
+}

model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c08431fe95c5ce2d0cdfd16e42e78763799f620c09ad51d16c141ffd67083d9c
+size 406487497

model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67740df7ede9a53e50d6e29c6a5c0d6c862f1876c22545d810515bad3ae17bb1
-size 455481275
+oid sha256:04ec04fdf7db74b4fed7f4b52f52e04dec4dbad9e4d88d2d178f334079a29fde
+size 455409152

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d70a9c7138318dbdc487fb9bcaa1b86ae9a41f6132d823a257e8d4a5fb42636
+size 384859388

sw_jax_cv_config.json

@@ -0,0 +1,19 @@
+{
+    "image_size": 448,
+    "model_name": "swinv2_base",
+    "model_args": {
+        "image_size": 448,
+        "patch_size": 4,
+        "in_chans": 3,
+        "num_classes": 9083,
+        "embed_dim": 128,
+        "window_size": 14,
+        "mlp_ratio": 4.0,
+        "qkv_bias": true,
+        "drop_rate": 0.0,
+        "attn_drop_rate": 0.0,
+        "drop_path_rate": 0.1,
+        "patch_norm": true,
+        "layer_norm_eps": 1e-05
+    }
+}