Add model

config.json

@@ -2,6 +2,9 @@
   "act_decoder": "relu",
   "act_encoder": "silu",
   "act_presnet": "relu",
+  "architectures": [
+    "RTDetrForObjectDetection"
+  ],
   "aux_loss": true,
   "block_nums": [
     3,
@@ -15,22 +18,28 @@
   "dim_feedforward": 1024,
   "dropout": 0.0,
   "enc_act": "gelu",
+  "eos_coefficient": 0.1,
   "eps": 0.01,
   "eval_idx": -1,
   "eval_size": null,
-  "eval_spatial_size": null,
+  "eval_spatial_size": [
+    640,
+    640
+  ],
   "expansion": 1.0,
   "feat_channels": [
-    512,
-    1024,
-    2048
+    256,
+    256,
+    256
   ],
   "feat_strides": [
     8,
     16,
     32
   ],
-  "freeze_at": -1,
+  "focal_loss_alpha": 0.75,
+  "focal_loss_gamma": 2.0,
+  "freeze_at": 0,
   "freeze_norm": true,
   "hidden_dim": 256,
   "in_channels": [
@@ -42,6 +51,11 @@
   "is_encoder_decoder": true,
   "label_noise_ratio": 0.5,
   "learnt_init_query": false,
+  "matcher_alpha": 0.25,
+  "matcher_bbox_cost": 5,
+  "matcher_class_cost": 2,
+  "matcher_gamma": 2.0,
+  "matcher_giou_cost": 2,
   "model_type": "rt_detr",
   "num_classes": 80,
   "num_decoder_layers": 6,
@@ -56,14 +70,18 @@
   "position_embed_type": "sine",
   "pretrained": true,
   "return_idx": [
-    0,
     1,
     2,
     3
   ],
+  "torch_dtype": "float32",
   "transformers_version": "4.35.0.dev0",
   "use_encoder_idx": [
     2
   ],
-  "variant": "d"
+  "use_focal_loss": true,
+  "variant": "d",
+  "weight_loss_bbox": 5,
+  "weight_loss_giou": 2,
+  "weight_loss_vfl": 1
 }

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb36532d579033ddb231c9d884a86477ce1574005c35d2976a48de3ddb36641b
+size 172334605