update big model backbone

ppcls/arch/backbone/legendary_models/open_clip/__init__.py

@@ -0,0 +1,96 @@
+import os
+import wget
+import paddle
+from .model_CLIP import CLIP, LaCLIP
+from paddle.vision.transforms import Compose, Resize, CenterCrop, ToTensor, Normalize
+from .unicom import load_model
+
+
+def clip_rn50():
+    model = CLIP(
+        embed_dim=1024,
+        image_resolution=224,
+        vision_layers=(3, 4, 6, 3),
+        vision_width=64,
+        vision_patch_size=None,
+        vision_head=64,
+        context_length=77,
+        vocab_size=49408,
+        transformer_width=512,
+        transformer_heads=8,
+        transformer_layers=12)
+    return model
+
+
+def clip_vit_b_32():
+    model = CLIP(
+        embed_dim=512,
+        image_resolution=224,
+        vision_layers=12,
+        vision_width=768,
+        vision_patch_size=32,
+        vision_head=64,
+        context_length=77,
+        vocab_size=49408,
+        transformer_width=512,
+        transformer_heads=8,
+        transformer_layers=12)
+    return model
+
+
+def clip_vit_b_16():
+    model = CLIP(
+        embed_dim=512,
+        image_resolution=224,
+        vision_layers=12,
+        vision_width=768,
+        vision_patch_size=16,
+        vision_head=64,
+        context_length=77,
+        vocab_size=49408,
+        transformer_width=512,
+        transformer_heads=8,
+        transformer_layers=12)
+    return model
+
+
+def Laclip_vit_b_32():
+    model = LaCLIP(
+        embed_dim=512,
+        image_resolution=224,
+        vision_layers=12,
+        vision_width=768,
+        vision_patch_size=32,
+        vision_head=64,
+        context_length=77,
+        vocab_size=49408,
+        transformer_width=512,
+        transformer_heads=8,
+        transformer_layers=12)
+    return model
+
+
+def Laclip_vit_b_16():
+    model = LaCLIP(
+        embed_dim=512,
+        image_resolution=224,
+        vision_layers=12,
+        vision_width=768,
+        vision_patch_size=16,
+        vision_head=64,
+        context_length=77,
+        vocab_size=49408,
+        transformer_width=512,
+        transformer_heads=8,
+        transformer_layers=12)
+    return model
+
+
+def unicom_vit_b_32():
+    model = load_model("ViT-B/32")
+    return model
+
+
+def Laclip_vit_b_16():
+    model = load_model("ViT-B/16")
+    return model