added 3rd model

app.py

@@ -9,6 +9,7 @@ from huggingface_hub import login, HfFileSystem, snapshot_download, HfApi, creat
 
 from app_train import create_training_demo
 from sdxl.app_inference import create_inference_demo
+from depthgltf.app_visualisation import create_visual_demo 
 
 css="""
 #col-container {max-width: 780px; margin-left: auto; margin-right: auto;}
@@ -86,9 +87,7 @@ with gr.Blocks(css=css) as demo:
     with gr.Tab("Generation"):
         create_inference_demo()
     with gr.Tab("Visualisation"):
-        gr.Markdown('''
-            - You can use this tab to upload models later if you choose not to upload models in training time or if upload in training time failed.
-            ''')
+        create_visual_demo(); 
 
 
 demo.queue(max_size=1).launch(debug=True, share=True)

depthgltf/.gitattributes

@@ -0,0 +1,27 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zstandard filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

depthgltf/.gitignore

@@ -0,0 +1,46 @@
+# Python build
+.eggs/
+gradio.egg-info/*
+!gradio.egg-info/requires.txt
+!gradio.egg-info/PKG-INFO
+dist/
+*.pyc
+__pycache__/
+*.py[cod]
+*$py.class
+build/
+
+# JS build
+gradio/templates/frontend
+# Secrets
+.env
+
+# Gradio run artifacts
+*.db
+*.sqlite3
+gradio/launches.json
+flagged/
+gradio_cached_examples/
+
+# Tests
+.coverage
+coverage.xml
+test.txt
+
+# Demos
+demo/tmp.zip
+demo/files/*.avi
+demo/files/*.mp4
+
+# Etc
+.idea/*
+.DS_Store
+*.bak
+workspace.code-workspace
+*.h5
+.vscode/
+
+# log files
+.pnpm-debug.log
+venv/
+*.db-journal

depthgltf/README.md

@@ -0,0 +1,12 @@
+---
+title: Dpt Depth Estimation + 3D
+emoji: ⚡
+colorFrom: blue
+colorTo: red
+sdk: gradio
+sdk_version: 3.0b8
+app_file: app.py
+pinned: false
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces#reference

depthgltf/app_visualisations.py

@@ -0,0 +1,119 @@
+import gradio as gr
+from transformers import DPTFeatureExtractor, DPTForDepthEstimation
+import torch
+import numpy as np
+from PIL import Image
+import open3d as o3d
+from pathlib import Path
+import os
+
+feature_extractor = DPTFeatureExtractor.from_pretrained("Intel/dpt-large")
+model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large")
+
+
+def process_image(image_path):
+    image_path = Path(image_path)
+    image_raw = Image.open(image_path)
+    image = image_raw.resize(
+        (800, int(800 * image_raw.size[1] / image_raw.size[0])),
+        Image.Resampling.LANCZOS)
+
+    # prepare image for the model
+    encoding = feature_extractor(image, return_tensors="pt")
+
+    # forward pass
+    with torch.no_grad():
+        outputs = model(**encoding)
+        predicted_depth = outputs.predicted_depth
+
+    # interpolate to original size
+    prediction = torch.nn.functional.interpolate(
+        predicted_depth.unsqueeze(1),
+        size=image.size[::-1],
+        mode="bicubic",
+        align_corners=False,
+    ).squeeze()
+    output = prediction.cpu().numpy()
+    depth_image = (output * 255 / np.max(output)).astype('uint8')
+    try:
+        gltf_path = create_3d_obj(np.array(image), depth_image, image_path)
+        img = Image.fromarray(depth_image)
+        return [img, gltf_path, gltf_path]
+    except Exception as e:
+        gltf_path = create_3d_obj(
+            np.array(image), depth_image, image_path, depth=8)
+        img = Image.fromarray(depth_image)
+        return [img, gltf_path, gltf_path]
+    except:
+        print("Error reconstructing 3D model")
+        raise Exception("Error reconstructing 3D model")
+
+
+def create_3d_obj(rgb_image, depth_image, image_path, depth=10):
+    depth_o3d = o3d.geometry.Image(depth_image)
+    image_o3d = o3d.geometry.Image(rgb_image)
+    rgbd_image = o3d.geometry.RGBDImage.create_from_color_and_depth(
+        image_o3d, depth_o3d, convert_rgb_to_intensity=False)
+    w = int(depth_image.shape[1])
+    h = int(depth_image.shape[0])
+
+    camera_intrinsic = o3d.camera.PinholeCameraIntrinsic()
+    camera_intrinsic.set_intrinsics(w, h, 500, 500, w/2, h/2)
+
+    pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
+        rgbd_image, camera_intrinsic)
+
+    print('normals')
+    pcd.normals = o3d.utility.Vector3dVector(
+        np.zeros((1, 3)))  # invalidate existing normals
+    pcd.estimate_normals(
+        search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.01, max_nn=30))
+    pcd.orient_normals_towards_camera_location(
+        camera_location=np.array([0., 0., 1000.]))
+    pcd.transform([[1, 0, 0, 0],
+                   [0, -1, 0, 0],
+                   [0, 0, -1, 0],
+                   [0, 0, 0, 1]])
+    pcd.transform([[-1, 0, 0, 0],
+                   [0, 1, 0, 0],
+                   [0, 0, 1, 0],
+                   [0, 0, 0, 1]])
+
+    print('run Poisson surface reconstruction')
+    with o3d.utility.VerbosityContextManager(o3d.utility.VerbosityLevel.Debug) as cm:
+        mesh_raw, densities = o3d.geometry.TriangleMesh.create_from_point_cloud_poisson(
+            pcd, depth=depth, width=0, scale=1.1, linear_fit=True)
+
+    voxel_size = max(mesh_raw.get_max_bound() - mesh_raw.get_min_bound()) / 256
+    print(f'voxel_size = {voxel_size:e}')
+    mesh = mesh_raw.simplify_vertex_clustering(
+        voxel_size=voxel_size,
+        contraction=o3d.geometry.SimplificationContraction.Average)
+
+    # vertices_to_remove = densities < np.quantile(densities, 0.001)
+    # mesh.remove_vertices_by_mask(vertices_to_remove)
+    bbox = pcd.get_axis_aligned_bounding_box()
+    mesh_crop = mesh.crop(bbox)
+    gltf_path = f'./{image_path.stem}.gltf'
+    o3d.io.write_triangle_mesh(
+        gltf_path, mesh_crop, write_triangle_uvs=True)
+    return gltf_path
+
+
+title = "Demo: zero-shot depth estimation with DPT + 3D Point Cloud"
+description = "This demo is a variation from the original <a href='https://huggingface.co/spaces/nielsr/dpt-depth-estimation' target='_blank'>DPT Demo</a>. It uses the DPT model to predict the depth of an image and then uses 3D Point Cloud to create a 3D object."
+examples = [["examples/" + img] for img in os.listdir("examples/")]
+
+iface = gr.Interface(fn=process_image,
+                     inputs=[gr.Image(
+                         type="filepath", label="Input Image")],
+                     outputs=[gr.Image(label="predicted depth", type="pil"),
+                              gr.Model3D(label="3d mesh reconstruction", clear_color=[
+                                                 1.0, 1.0, 1.0, 1.0]),
+                              gr.File(label="3d gLTF")],
+                     title=title,
+                     description=description,
+                     examples=examples,
+                     allow_flagging="never",
+                     cache_examples=False)
+iface.launch(debug=True, enable_queue=False, share=True)

depthgltf/packages.txt

@@ -0,0 +1 @@
+libgl1-mesa-glx

depthgltf/requirements.txt

@@ -0,0 +1,7 @@
+torch
+git+https://github.com/nielsrogge/transformers.git@add_dpt_redesign#egg=transformers
+numpy
+Pillow
+gradio
+jinja2
+open3d