Refactor/docker pipelines

Steps.ipynb

@@ -0,0 +1,205 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "749ffe22-1a4a-4cba-8004-8870695ebe00",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "import argparse\n",
+    "import copy\n",
+    "import os\n",
+    "import time\n",
+    "\n",
+    "import cv2\n",
+    "import imageio\n",
+    "import numpy as np\n",
+    "import torch\n",
+    "import vispy\n",
+    "import yaml\n",
+    "from tqdm import tqdm\n",
+    "\n",
+    "import sys\n",
+    "sys.path.append(\"/art_pipelines/libs/3d-photo-inpainting/\")\n",
+    "\n",
+    "import MiDaS.MiDaS_utils as MiDaS_utils\n",
+    "from MiDaS.monodepth_net import MonoDepthNet\n",
+    "from MiDaS.run import run_depth\n",
+    "from bilateral_filtering import sparse_bilateral_filtering\n",
+    "from boostmonodepth_utils import run_boostmonodepth, resize_depth,replace_ext, write_depth\n",
+    "from mesh import output_3d_photo, read_ply, write_ply\n",
+    "from networks import Inpaint_Color_Net, Inpaint_Depth_Net, Inpaint_Edge_Net\n",
+    "from utils import get_MiDaS_samples, read_MiDaS_depth"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "b2d6aa81-6bc0-411f-8d47-d82b8877eb54",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "workspace_dir = \"/workspace\"\n",
+    "\n",
+    "depth_dir = f\"{workspace_dir}/depth_folder\"\n",
+    "\n",
+    "os.makedirs(f\"{workspace_dir}/mesh_folder\", exist_ok=True)\n",
+    "os.makedirs(f\"{workspace_dir}/video_folder\", exist_ok=True)\n",
+    "os.makedirs(depth_dir, exist_ok=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "2e315181-98ab-4899-86b2-e8b827a24304",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "import pydantic\n",
+    "from typing import List, Dict, Any, Union"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "dd432633-ca78-4810-ac04-2e0fee06f6ab",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "class MidasConfig(pydantic.BaseModel):\n",
+    "    traj_types: List[str] = ['straight-line']\n",
+    "    x_shift_range: List[float] = [-0.04]\n",
+    "    y_shift_range: List[float] = [0.00]\n",
+    "    z_shift_range: List[float] = [0.00]\n",
+    "    video_postfix: List[Any] = ['left']\n",
+    "    depth_format: str = '.npy'\n",
+    "    num_frames: int = 4"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "77cb0204-eeb6-446e-8ae3-67e9c82ca451",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "source_dir = \"/workspace/source_images\"\n",
+    "\n",
+    "config = MidasConfig().dict()\n",
+    "\n",
+    "sample_list = get_MiDaS_samples(\n",
+    "    image_folder=source_dir,\n",
+    "    depth_folder=depth_dir,\n",
+    "    config=config,\n",
+    "    specific='', \n",
+    "    aft_certain=None\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "0406700c-9c90-462e-a15b-83277e886315",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "BOOST_BASE = \"/art_pipelines/libs/3d-photo-inpainting/BoostingMonocularDepth/\"\n",
+    "\n",
+    "os.system(\n",
+    "    f\"cd {BOOST_BASE} && python run.py --Final --data_dir {source_dir}/  --output_dir {depth_dir} --depthNet 0\"\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e766eb2c-b09b-422d-87cc-85eb0737486d",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "import numpy"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "3a7ae8f7-f177-4454-93bc-7bf4238e1917",
+   "metadata": {
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "image_names = [sample[\"ref_img_fi\"] for sample in sample_list]\n",
+    "\n",
+    "for i, image_name in enumerate(image_names):\n",
+    "    img = imageio.imread(image_name)\n",
+    "    H, W = img.shape[:2]\n",
+    "    scale = 640.0 / max(H, W)\n",
+    "\n",
+    "    # resize and save depth\n",
+    "    target_height, target_width = int(round(H * scale)), int(round(W * scale))\n",
+    "    depth = imageio.imread(os.path.join(depth_dir, image_name))\n",
+    "    depth = np.array(depth).astype(np.float32)\n",
+    "    depth = resize_depth(depth, target_width, target_height)\n",
+    "    output_path = os.path.join(depth_dir, replace_ext(os.path.basename(image_name), \"npy\"))\n",
+    "    print(f\"Writing depth .npy to {output_path}\")\n",
+    "    np.save(output_path, depth / 32768.0 - 1.0)\n",
+    "    output_path = os.path.join(depth_dir, os.path.basename(image_name))\n",
+    "    print(f\"Writing depth to {output_path}\")\n",
+    "    write_depth(output_path, depth)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.7.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

boostmonodepth_utils.py

@@ -1,14 +1,16 @@
+import glob
 import os
+
 import cv2
-import glob
-import numpy as np
 import imageio
+import numpy as np
 from MiDaS.MiDaS_utils import write_depth
 
-BOOST_BASE = 'BoostingMonocularDepth'
+BOOST_BASE = "BoostingMonocularDepth"
+
+BOOST_INPUTS = "inputs"
+BOOST_OUTPUTS = "outputs"
 
-BOOST_INPUTS = 'inputs'
-BOOST_OUTPUTS = 'outputs'
 
 def run_boostmonodepth(img_names, src_folder, depth_folder):
 
@@ -21,37 +23,41 @@ def run_boostmonodepth(img_names, src_folder, depth_folder):
 
     tgt_names = []
     for img_name in img_names:
-        base_name = os.path.basename(img_name)
-        tgt_name = os.path.join(BOOST_BASE, BOOST_INPUTS, base_name)
-        os.system(f'cp {img_name} {tgt_name}')
+        base_name = os.path.basename(replace_ext(img_name, ".png"))
+        input_file = os.path.join(BOOST_BASE, BOOST_INPUTS, base_name)
+        os.system(f"cp {img_name} {input_file}")
+        tgt_names.append(base_name)
 
         # keep only the file name here.
         # they save all depth as .png file
-        tgt_names.append(os.path.basename(tgt_name).replace('.jpg', '.png'))
 
-    os.system(f'cd {BOOST_BASE} && python run.py --Final --data_dir {BOOST_INPUTS}/  --output_dir {BOOST_OUTPUTS} --depthNet 0')
+    os.system(
+        f"cd {BOOST_BASE} && python run.py --Final --data_dir {BOOST_INPUTS}/  --output_dir {BOOST_OUTPUTS} --depthNet 0"
+    )
 
     for i, (img_name, tgt_name) in enumerate(zip(img_names, tgt_names)):
         img = imageio.imread(img_name)
         H, W = img.shape[:2]
-        scale = 640. / max(H, W)
+        scale = 640.0 / max(H, W)
 
         # resize and save depth
         target_height, target_width = int(round(H * scale)), int(round(W * scale))
         depth = imageio.imread(os.path.join(BOOST_BASE, BOOST_OUTPUTS, tgt_name))
         depth = np.array(depth).astype(np.float32)
         depth = resize_depth(depth, target_width, target_height)
-        np.save(os.path.join(depth_folder, tgt_name.replace('.png', '.npy')), depth / 32768. - 1.)
-        write_depth(os.path.join(depth_folder, tgt_name.replace('.png', '')), depth)
+        np.save(os.path.join(depth_folder, replace_ext(tgt_name, ".npy")), depth / 32768.0 - 1.0)
+        write_depth(os.path.join(depth_folder, replace_ext(tgt_name, "")), depth)
 
-def clean_folder(folder, img_exts=['.png', '.jpg', '.npy']):
 
+def clean_folder(folder, img_exts=None):
+    img_exts = img_exts or [".png", ".jpg", ".npy"]
     for img_ext in img_exts:
-        paths_to_check = os.path.join(folder, f'*{img_ext}')
+        paths_to_check = os.path.join(folder, f"*{img_ext}")
         if len(glob.glob(paths_to_check)) == 0:
             continue
         print(paths_to_check)
-        os.system(f'rm {paths_to_check}')
+        os.system(f"rm {paths_to_check}")
+
 
 def resize_depth(depth, width, height):
     """Resize numpy (or image read by imageio) depth map
@@ -66,3 +72,8 @@ def resize_depth(depth, width, height):
     """
     depth = cv2.blur(depth, (3, 3))
     return cv2.resize(depth, (width, height), interpolation=cv2.INTER_AREA)
+
+
+def replace_ext(filename: str, extension: str):
+    extension = extension.replace(".", "")
+    return f"{os.path.splitext(filename)[0]}.{extension}"

docker-compose.yaml

@@ -0,0 +1,27 @@
+version: '2.3'
+
+services:
+  vispy:
+    build:
+      context: .
+      dockerfile: ./vispy.Dockerfile
+    #image: pytorch/pytorch:latest
+    container_name: fastapi
+    environment:
+      PYTHONPATH: /art_pipelines
+    volumes:
+      - ../:/art_pipelines
+      - ../../server:/workspace/server:ro
+      - ../../temp:/workspace/temp
+    #entrypoint: "dir ./"
+    entrypoint: "bash -c 'jupyter lab --no-browser --ip=0.0.0.0 --port=8893 --allow-root'"
+    ports:
+      - "8893:8893"
+    # enable gpu for linux (runtime: nvidia) or windows (deploy:...)
+    #runtime: nvidia
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - capabilities:
+                - gpu