Add config from convert_d_fine_original_pytorch_checkpoint_to_hf.py

config.json

@@ -1,10 +1,8 @@
 {
+  "_attn_implementation_autoset": true,
   "activation_dropout": 0.0,
   "activation_function": "silu",
   "anchor_image_size": null,
-  "architectures": [
-    "DFineForObjectDetection"
-  ],
   "attention_dropout": 0.0,
   "auxiliary_loss": true,
   "backbone": null,
@@ -21,6 +19,29 @@
       2,
       3,
       4
+    ],
+    "stage_in_channels": [
+      64,
+      128,
+      512,
+      1024
+    ],
+    "stage_mid_channels": [
+      64,
+      128,
+      256,
+      512
+    ],
+    "stage_num_blocks": [
+      1,
+      2,
+      5,
+      2
+    ],
+    "stem_channels": [
+      3,
+      32,
+      64
     ]
   },
   "backbone_kwargs": null,
@@ -31,13 +52,17 @@
   "decoder_attention_heads": 8,
   "decoder_ffn_dim": 1024,
   "decoder_in_channels": [
-    256,
-    256,
-    256
+    384,
+    384,
+    384
   ],
   "decoder_layers": 6,
   "decoder_method": "default",
-  "decoder_n_points": 4,
+  "decoder_n_points": [
+    3,
+    6,
+    3
+  ],
   "decoder_offset_scale": 0.5,
   "depth_mult": 1.0,
   "disable_custom_kernels": true,
@@ -47,8 +72,8 @@
   ],
   "encoder_activation_function": "gelu",
   "encoder_attention_heads": 8,
-  "encoder_ffn_dim": 1024,
-  "encoder_hidden_dim": 256,
+  "encoder_ffn_dim": 2048,
+  "encoder_hidden_dim": 384,
   "encoder_in_channels": [
     512,
     1024,
@@ -254,7 +279,6 @@
   "positional_encoding_temperature": 10000,
   "reg_scale": 4.0,
   "top_prob_values": 4,
-  "torch_dtype": "float32",
   "transformers_version": "4.50.0.dev0",
   "use_focal_loss": true,
   "use_pretrained_backbone": false,