new

.idea/.gitignore

@@ -0,0 +1,8 @@
+# 默认忽略的文件
+/shelf/
+/workspace.xml
+# 基于编辑器的 HTTP 客户端请求
+/httpRequests/
+# Datasource local storage ignored files
+/dataSources/
+/dataSources.local.xml

config.json

@@ -0,0 +1,114 @@
+{
+  "add_projection": false,
+  "architectures": [
+    "ZoeDepthForDepthEstimation"
+  ],
+  "attractor_alpha": 1000,
+  "attractor_gamma": 2,
+  "attractor_kind": "mean",
+  "backbone": null,
+  "backbone_config": {
+    "hidden_size": 1024,
+    "image_size": 384,
+    "intermediate_size": 4096,
+    "model_type": "beit",
+    "num_attention_heads": 16,
+    "num_hidden_layers": 24,
+    "out_features": [
+      "stage6",
+      "stage12",
+      "stage18",
+      "stage24"
+    ],
+    "out_indices": [
+      6,
+      12,
+      18,
+      24
+    ],
+    "reshape_hidden_states": false,
+    "stage_names": [
+      "stem",
+      "stage1",
+      "stage2",
+      "stage3",
+      "stage4",
+      "stage5",
+      "stage6",
+      "stage7",
+      "stage8",
+      "stage9",
+      "stage10",
+      "stage11",
+      "stage12",
+      "stage13",
+      "stage14",
+      "stage15",
+      "stage16",
+      "stage17",
+      "stage18",
+      "stage19",
+      "stage20",
+      "stage21",
+      "stage22",
+      "stage23",
+      "stage24"
+    ],
+    "use_relative_position_bias": true
+  },
+  "backbone_hidden_size": 1024,
+  "batch_norm_eps": 1e-05,
+  "bin_centers_type": "softplus",
+  "bin_configurations": [
+    {
+      "max_depth": 10.0,
+      "min_depth": 0.001,
+      "n_bins": 64,
+      "name": "nyu"
+    },
+    {
+      "max_depth": 80.0,
+      "min_depth": 0.001,
+      "n_bins": 64,
+      "name": "kitti"
+    }
+  ],
+  "bin_embedding_dim": 128,
+  "bottleneck_features": 256,
+  "fusion_hidden_size": 256,
+  "head_in_index": -1,
+  "hidden_act": "gelu",
+  "initializer_range": 0.02,
+  "max_temp": 50.0,
+  "min_temp": 0.0212,
+  "model_type": "zoedepth",
+  "neck_hidden_sizes": [
+    256,
+    512,
+    1024,
+    1024
+  ],
+  "num_attractors": [
+    16,
+    8,
+    4,
+    1
+  ],
+  "num_patch_transformer_layers": 4,
+  "num_relative_features": 32,
+  "patch_transformer_hidden_size": 128,
+  "patch_transformer_intermediate_size": 1024,
+  "patch_transformer_num_attention_heads": 4,
+  "readout_type": "project",
+  "reassemble_factors": [
+    4,
+    2,
+    1,
+    0.5
+  ],
+  "torch_dtype": "float32",
+  "transformers_version": "4.42.0.dev0",
+  "use_batch_norm_in_fusion_residual": false,
+  "use_bias_in_fusion_residual": null,
+  "use_pretrained_backbone": false
+}

handler.py

@@ -0,0 +1,56 @@
+import base64
+import io
+import json
+from typing import Dict, Any
+from PIL import Image
+from transformers import pipeline
+
+class EndpointHandler:
+    """
+    Custom handler for the ZoeDepth model, fully compliant with the latest
+    Hugging Face Inference Endpoints documentation.
+    The final result is serialized into a single JSON string.
+    """
+    def __init__(self, path=""):
+        # Initialize the pipeline for depth-estimation
+        self.pipe = pipeline(task="depth-estimation", model=path)
+        print("Depth estimation pipeline initialized successfully.")
+
+    def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
+        """
+        This method is called for every API request.
+        
+        Args:
+            data (Dict): The input data dictionary. Expects "inputs" key with image bytes.
+        
+        Returns:
+            Dict[str, str]: A dictionary with a single key "generated_text", 
+                            containing a JSON string of the results.
+        """
+        # Get image bytes from the request
+        inputs = data.pop("inputs", data)
+        image = Image.open(io.BytesIO(inputs))
+
+        # Pass the image to the pipeline
+        prediction = self.pipe(image)
+        
+        # Extract raw depth data and visual map
+        raw_depth_tensor = prediction["predicted_depth"]
+        raw_depth_data = raw_depth_tensor.cpu().tolist()
+
+        visual_map_image = prediction["depth"]
+        buffered = io.BytesIO()
+        visual_map_image.save(buffered, format="PNG")
+        visual_map_base64 = base64.b64encode(buffered.getvalue()).decode("utf-8")
+
+        # Create a dictionary to hold all results
+        results = {
+            "raw_depth_data": raw_depth_data,
+            "visual_depth_map": f"data:image/png;base64,{visual_map_base64}"
+        }
+
+        # Serialize the entire results dictionary into a JSON string
+        json_output_string = json.dumps(results)
+
+        # Return the final dictionary in the required format
+        return {"generated_text": json_output_string}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5494fa0938f18d71e215e245472470c3aefebd7b434abd89750e5ae4008e2dc
+size 1380374404

preprocessor_config.json

@@ -0,0 +1,42 @@
+{
+  "_valid_processor_keys": [
+    "images",
+    "do_resize",
+    "size",
+    "keep_aspect_ratio",
+    "ensure_multiple_of",
+    "resample",
+    "do_rescale",
+    "rescale_factor",
+    "do_normalize",
+    "image_mean",
+    "image_std",
+    "do_pad",
+    "return_tensors",
+    "data_format",
+    "input_data_format"
+  ],
+  "do_normalize": true,
+  "do_pad": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "ensure_multiple_of": 32,
+  "image_mean": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "image_processor_type": "ZoeDepthImageProcessor",
+  "image_std": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "keep_aspect_ratio": true,
+  "resample": 2,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "height": 384,
+    "width": 512
+  }
+}