Upload folder using huggingface_hub

README.md

@@ -1,20 +1,4 @@
----
-license: apache-2.0
----
-# Handcrafted solution example for the S23DR competition
-
-This repo provides a minimalistic example of a wireframe estimation submission to S23DR competition.
-We recommend you take a look at [this example](https://github.com/s23dr/hoho2025/blob/main/hoho2025/example_solutions.py), for detailed code of this submission. It also provides useful I/O and visualization functions.
-
-This example seeks to simply provide minimal code which succeeds at reading the dataset and producing a solution (in this case two vertices at the origin and edge of zero length connecting them).
-
-`script.py` - is the main file which is run by the competition space. It should produce `submission.parquet` as the result of the run. Please see the additional comments in the `script.py` file. 
-
-
-# How to submit
-
-Use the notebook [example_notebook.ipynb](example_notebook.ipynb)
-
-
+# Handcrafted Submission 2025-1
 
+This repo contains a submission to the [S23DR Challenge](https://huggingface.co/spaces/usm3d/S23DR) (part of the [USM3D](https://usm3d.github.io/) workshop at CVPR2025). It was prepared by [bulkobubulko](https://huggingface.co/bulkobubulko).
 

script.py

@@ -15,7 +15,7 @@ from joblib import Parallel, delayed
 import os
 import json
 import gc
-from hoho2025.example_solutions import predict_wireframe
+from tools2025.hoho2025.hoho2025.example_solutions import predict_wireframe
 # check the https://github.com/s23dr/hoho2025/blob/main/hoho2025/example_solutions.py for the example solution
 
 def empty_solution():

tools2025/.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt 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
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz 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
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl 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
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* 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
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

tools2025/.gitignore

@@ -0,0 +1,3 @@
+.DS_Store
+__pycache__
+hoho.egg-info/

tools2025/LICENSE.txt

@@ -0,0 +1,13 @@
+Copyright 2024 Jack Langerman & Dmytro Mishkin
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.

tools2025/README.md

@@ -0,0 +1,81 @@
+---
+license: apache-2.0
+---
+# HoHo2025 Tools
+
+Tools and utilities for the [S23DR-2025 competition](https://huggingface.co/spaces/usm3d/S23DR2025) and [HoHo25k Dataset](https://huggingface.co/datasets/usm3d/hoho25k)
+
+## Installation 
+
+```bash
+pip install hoho2025
+```
+
+### pip install over http
+```bash
+pip install git+http://hf.co/usm3d/tools2025.git    
+```
+
+or editable 
+```bash
+git clone http://hf.co/usm3d/tools2025 
+cd tools2025
+pip install -e .
+```
+
+### Usage example
+
+```python
+from datasets import load_dataset
+from hoho2025.vis import plot_all_modalities
+from hoho2025.viz3d import *
+
+def read_colmap_rec(colmap_data):
+    import pycolmap
+    import tempfile,zipfile
+    import io
+    with tempfile.TemporaryDirectory() as tmpdir:
+        with zipfile.ZipFile(io.BytesIO(colmap_data), "r") as zf:
+            zf.extractall(tmpdir)  # unpacks cameras.txt, images.txt, etc. to tmpdir
+        # Now parse with pycolmap
+        rec = pycolmap.Reconstruction(tmpdir)
+        return rec
+
+ds = load_dataset("usm3d/hoho25k", streaming=True, trust_remote_code=True)
+for a in ds['train']:
+    break
+
+fig, ax = plot_all_modalities(a)
+
+## Now 3d
+
+fig3d = init_figure()
+plot_reconstruction(fig3d, read_colmap_rec(a['colmap_binary']))
+plot_wireframe(fig3d, a['wf_vertices'], a['wf_edges'], a['wf_classifications'])
+plot_bpo_cameras_from_entry(fig3d, a)
+fig3d
+```
+
+## Example wireframe estimation 
+
+Look in [hoho2025/example_solution.py](hoho2025/example_solution.py)
+
+```python
+from hoho2025.example_solutions import predict_wireframe
+pred_vertices, pred_connections = predict_wireframe(a)
+
+fig3d = init_figure()
+plot_reconstruction(fig3d, read_colmap_rec(a['colmap_binary']))
+plot_wireframe(fig3d, pred_vertices, pred_connections, color='rgb(0, 0, 255)')
+fig3d
+```
+
+
+And to get the metric
+
+```python
+from hoho2025.metric_helper import hss
+
+score = hss(pred_vertices, pred_connections, a['wf_vertices'], a['wf_edges'], vert_thresh=0.5, edge_thresh=0.5)
+print (score)
+```

tools2025/hoho2025/__init__.py

@@ -0,0 +1,25 @@
+from .hoho import *
+from . import vis
+
+import importlib
+import sys
+
+class LazyLoadModule:
+    def __init__(self, module_name):
+        self.module_name = module_name
+        self.module = None
+
+    def __getattribute__(self, attr):
+        if attr == 'module_name' or attr == 'module':
+            return super().__getattribute__(attr)
+
+        if self.module is None:
+            self.module = importlib.import_module(f'hoho.{self.module_name}')
+            sys.modules[self.module_name] = self.module
+
+        return getattr(self.module, attr)
+    
+try:
+    import viz3d
+except ImportError:
+    viz3d = LazyLoadModule('viz3d')

tools2025/hoho2025/color_mappings.py

@@ -0,0 +1,209 @@
+gestalt_color_mapping = {
+    "unclassified": (215, 62, 138),
+    "apex": (235, 88, 48),
+    "eave_end_point": (248, 130, 228),
+    "flashing_end_point": (71, 11, 161),
+    "ridge": (214, 251, 248),
+    "rake": (13, 94, 47),
+    "eave": (54, 243, 63),
+    "post": (187, 123, 236),
+    "ground_line": (136, 206, 14),
+    "flashing": (162, 162, 32),
+    "step_flashing": (169, 255, 219),
+    "hip": (8, 89, 52),
+    "valley": (85, 27, 65),
+    "roof": (215, 232, 179),
+    "door": (110, 52, 23),
+    "garage": (50, 233, 171),
+    "window": (230, 249, 40),
+    "shutter": (122, 4, 233),
+    "fascia": (95, 230, 240),
+    "soffit": (2, 102, 197),
+    "horizontal_siding": (131, 88, 59),
+    "vertical_siding": (110, 187, 198),
+    "brick": (171, 252, 7),
+    "concrete": (32, 47, 246),
+    "other_wall": (112, 61, 240),
+    "trim": (151, 206, 58),
+    "unknown": (127, 127, 127),
+    "transition_line": (0,0,0),
+}
+
+ade20k_color_mapping = {
+    'wall': (120, 120, 120),
+    'building;edifice': (180, 120, 120),
+    'sky': (6, 230, 230),
+    'floor;flooring': (80, 50, 50),
+    'tree': (4, 200, 3),
+    'ceiling': (120, 120, 80),
+    'road;route': (140, 140, 140),
+    'bed': (204, 5, 255),
+    'windowpane;window': (230, 230, 230),
+    'grass': (4, 250, 7),
+    'cabinet': (224, 5, 255),
+    'sidewalk;pavement': (235, 255, 7),
+    'person;individual;someone;somebody;mortal;soul': (150, 5, 61),
+    'earth;ground': (120, 120, 70),
+    'door;double;door': (8, 255, 51),
+    'table': (255, 6, 82),
+    'mountain;mount': (143, 255, 140),
+    'plant;flora;plant;life': (204, 255, 4),
+    'curtain;drape;drapery;mantle;pall': (255, 51, 7),
+    'chair': (204, 70, 3),
+    'car;auto;automobile;machine;motorcar': (0, 102, 200),
+    'water': (61, 230, 250),
+    'painting;picture': (255, 6, 51),
+    'sofa;couch;lounge': (11, 102, 255),
+    'shelf': (255, 7, 71),
+    'house': (255, 9, 224),
+    'sea': (9, 7, 230),
+    'mirror': (220, 220, 220),
+    'rug;carpet;carpeting': (255, 9, 92),
+    'field': (112, 9, 255),
+    'armchair': (8, 255, 214),
+    'seat': (7, 255, 224),
+    'fence;fencing': (255, 184, 6),
+    'desk': (10, 255, 71),
+    'rock;stone': (255, 41, 10),
+    'wardrobe;closet;press': (7, 255, 255),
+    'lamp': (224, 255, 8),
+    'bathtub;bathing;tub;bath;tub': (102, 8, 255),
+    'railing;rail': (255, 61, 6),
+    'cushion': (255, 194, 7),
+    'base;pedestal;stand': (255, 122, 8),
+    'box': (0, 255, 20),
+    'column;pillar': (255, 8, 41),
+    'signboard;sign': (255, 5, 153),
+    'chest;of;drawers;chest;bureau;dresser': (6, 51, 255),
+    'counter': (235, 12, 255),
+    'sand': (160, 150, 20),
+    'sink': (0, 163, 255),
+    'skyscraper': (140, 140, 140),
+    'fireplace;hearth;open;fireplace': (250, 10, 15),
+    'refrigerator;icebox': (20, 255, 0),
+    'grandstand;covered;stand': (31, 255, 0),
+    'path': (255, 31, 0),
+    'stairs;steps': (255, 224, 0),
+    'runway': (153, 255, 0),
+    'case;display;case;showcase;vitrine': (0, 0, 255),
+    'pool;table;billiard;table;snooker;table': (255, 71, 0),
+    'pillow': (0, 235, 255),
+    'screen;door;screen': (0, 173, 255),
+    'stairway;staircase': (31, 0, 255),
+    'river': (11, 200, 200),
+    'bridge;span': (255 ,82, 0),
+    'bookcase': (0, 255, 245),
+    'blind;screen': (0, 61, 255),
+    'coffee;table;cocktail;table': (0, 255, 112),
+    'toilet;can;commode;crapper;pot;potty;stool;throne': (0, 255, 133),
+    'flower': (255, 0, 0),
+    'book': (255, 163, 0),
+    'hill': (255, 102, 0),
+    'bench': (194, 255, 0),
+    'countertop': (0, 143, 255),
+    'stove;kitchen;stove;range;kitchen;range;cooking;stove': (51, 255, 0),
+    'palm;palm;tree': (0, 82, 255),
+    'kitchen;island': (0, 255, 41),
+    'computer;computing;machine;computing;device;data;processor;electronic;computer;information;processing;system': (0, 255, 173),
+    'swivel;chair': (10, 0, 255),
+    'boat': (173, 255, 0),
+    'bar': (0, 255, 153),
+    'arcade;machine': (255, 92, 0),
+    'hovel;hut;hutch;shack;shanty': (255, 0, 255),
+    'bus;autobus;coach;charabanc;double-decker;jitney;motorbus;motorcoach;omnibus;passenger;vehicle': (255, 0, 245),
+    'towel': (255, 0, 102),
+    'light;light;source': (255, 173, 0),
+    'truck;motortruck': (255, 0, 20),
+    'tower': (255, 184, 184),
+    'chandelier;pendant;pendent': (0, 31, 255),
+    'awning;sunshade;sunblind': (0, 255, 61),
+    'streetlight;street;lamp': (0, 71, 255),
+    'booth;cubicle;stall;kiosk': (255, 0, 204),
+    'television;television;receiver;television;set;tv;tv;set;idiot;box;boob;tube;telly;goggle;box': (0, 255, 194),
+    'airplane;aeroplane;plane': (0, 255, 82),
+    'dirt;track': (0, 10, 255),
+    'apparel;wearing;apparel;dress;clothes': (0, 112, 255),
+    'pole': (51, 0, 255),
+    'land;ground;soil': (0, 194, 255),
+    'bannister;banister;balustrade;balusters;handrail': (0, 122, 255),
+    'escalator;moving;staircase;moving;stairway': (0, 255, 163),
+    'ottoman;pouf;pouffe;puff;hassock': (255, 153, 0),
+    'bottle': (0, 255, 10),
+    'buffet;counter;sideboard': (255, 112, 0),
+    'poster;posting;placard;notice;bill;card': (143, 255, 0),
+    'stage': (82, 0, 255),
+    'van': (163, 255, 0),
+    'ship': (255, 235, 0),
+    'fountain': (8, 184, 170),
+    'conveyer;belt;conveyor;belt;conveyer;conveyor;transporter': (133, 0, 255),
+    'canopy': (0, 255, 92),
+    'washer;automatic;washer;washing;machine': (184, 0, 255),
+    'plaything;toy': (255, 0, 31),
+    'swimming;pool;swimming;bath;natatorium': (0, 184, 255),
+    'stool': (0, 214, 255),
+    'barrel;cask': (255, 0, 112),
+    'basket;handbasket': (92, 255, 0),
+    'waterfall;falls': (0, 224, 255),
+    'tent;collapsible;shelter': (112, 224, 255),
+    'bag': (70, 184, 160),
+    'minibike;motorbike': (163, 0, 255),
+    'cradle': (153, 0, 255),
+    'oven': (71, 255, 0),
+    'ball': (255, 0, 163),
+    'food;solid;food': (255, 204, 0),
+    'step;stair': (255, 0, 143),
+    'tank;storage;tank': (0, 255, 235),
+    'trade;name;brand;name;brand;marque': (133, 255, 0),
+    'microwave;microwave;oven': (255, 0, 235),
+    'pot;flowerpot': (245, 0, 255),
+    'animal;animate;being;beast;brute;creature;fauna': (255, 0, 122),
+    'bicycle;bike;wheel;cycle': (255, 245, 0),
+    'lake': (10, 190, 212),
+    'dishwasher;dish;washer;dishwashing;machine': (214, 255, 0),
+    'screen;silver;screen;projection;screen': (0, 204, 255),
+    'blanket;cover': (20, 0, 255),
+    'sculpture': (255, 255, 0),
+    'hood;exhaust;hood': (0, 153, 255),
+    'sconce': (0, 41, 255),
+    'vase': (0, 255, 204),
+    'traffic;light;traffic;signal;stoplight': (41, 0, 255),
+    'tray': (41, 255, 0),
+    'ashcan;trash;can;garbage;can;wastebin;ash;bin;ash-bin;ashbin;dustbin;trash;barrel;trash;bin': (173, 0, 255),
+    'fan': (0, 245, 255),
+    'pier;wharf;wharfage;dock': (71, 0, 255),
+    'crt;screen': (122, 0, 255),
+    'plate': (0, 255, 184),
+    'monitor;monitoring;device': (0, 92, 255),
+    'bulletin;board;notice;board': (184, 255, 0),
+    'shower': (0, 133, 255),
+    'radiator': (255, 214, 0),
+    'glass;drinking;glass': (25, 194, 194),
+    'clock': (102, 255, 0),
+    'flag': (92, 0, 255),
+}
+
+
+EDGE_CLASSES = {'cornice_return': 0,
+                'cornice_strip': 1,
+                'eave': 2,
+                'flashing': 3,
+                'hip': 4,
+                'rake': 5,
+                'ridge': 6,
+                'step_flashing': 7,
+                'transition_line': 8,
+                'valley': 9}
+EDGE_CLASSES_BY_ID = {v: k for k, v in EDGE_CLASSES.items()}
+
+edge_color_mapping = {
+    'cornice_return': (215, 62, 138),
+    'cornice_strip': (235, 88, 48),
+    'eave':  (54, 243, 63),
+    "flashing": (162, 162, 32),
+    'hip': (8, 89, 52),
+    'rake': (13, 94, 47),
+    'ridge': (214, 251, 248),
+    "step_flashing": (169, 255, 219),
+    'transition_line': (200,0,50),
+    'valley': (85, 27, 65),
+}

tools2025/hoho2025/example_solutions.py

@@ -0,0 +1,715 @@
+# Description: This file contains the handcrafted solution for the task of wireframe reconstruction 
+import io
+import tempfile
+import zipfile
+from collections import defaultdict
+from typing import Tuple, List
+import cv2
+import numpy as np
+import pycolmap
+from PIL import Image as PImage
+from scipy.spatial.distance import cdist
+
+from hoho2025.color_mappings import ade20k_color_mapping, gestalt_color_mapping
+
+
+def empty_solution():
+    '''Return a minimal valid solution, i.e. 2 vertices and 1 edge.'''
+    return np.zeros((2,3)), [(0, 1)]
+    
+    
+def read_colmap_rec(colmap_data):
+    with tempfile.TemporaryDirectory() as tmpdir:
+        with zipfile.ZipFile(io.BytesIO(colmap_data), "r") as zf:
+            zf.extractall(tmpdir)  # unpacks cameras.txt, images.txt, etc. to tmpdir
+        # Now parse with pycolmap
+        rec = pycolmap.Reconstruction(tmpdir)
+        return rec
+
+def convert_entry_to_human_readable(entry):
+    out = {}
+    for k, v in entry.items():
+        if 'colmap' in k:
+            out[k] = read_colmap_rec(v)
+        elif k in ['wf_vertices', 'wf_edges', 'K', 'R', 't', 'depth']:
+            out[k] = np.array(v)
+        else:
+            out[k]=v
+    out['__key__'] = entry['order_id']
+    return out
+
+
+def get_house_mask(ade20k_seg):
+    """
+    Get a mask of the house in the ADE20K segmentation map.
+    """
+    house_classes_ade20k = [
+        'wall',
+        'house',
+        'building;edifice',
+        'door;double;door',
+        'windowpane;window',
+    ]
+    np_seg = np.array(ade20k_seg)
+    full_mask = np.zeros(np_seg.shape[:2], dtype=np.uint8)
+    for c in house_classes_ade20k:
+        color = np.array(ade20k_color_mapping[c])
+        mask = cv2.inRange(np_seg, color-0.5, color+0.5)
+        full_mask = np.logical_or(full_mask, mask)
+    return full_mask
+
+
+def point_to_segment_dist(pt, seg_p1, seg_p2):
+    """
+    Computes the Euclidean distance from pt to the line segment p1->p2.
+    pt, seg_p1, seg_p2: (x, y) as np.ndarray
+    """
+    # If both endpoints are the same, just return distance to one of them
+    if np.allclose(seg_p1, seg_p2):
+        return np.linalg.norm(pt - seg_p1)
+    seg_vec = seg_p2 - seg_p1
+    pt_vec = pt - seg_p1
+    seg_len2 = seg_vec.dot(seg_vec)
+    t = max(0, min(1, pt_vec.dot(seg_vec)/seg_len2))
+    proj = seg_p1 + t*seg_vec
+    return np.linalg.norm(pt - proj)
+
+
+def get_vertices_and_edges_from_segmentation(gest_seg_np, edge_th=25.0):
+    """
+    Identify apex and eave-end vertices, then detect lines for eave/ridge/rake/valley.
+    For each connected component, we do a line fit with cv2.fitLine, then measure
+    segment endpoints more robustly. We then associate apex points that are within
+    'edge_th' of the line segment. We record those apex–apex connections for edges
+    if at least 2 apexes lie near the same component line.
+    """
+    #--------------------------------------------------------------------------------
+    # Step A: Collect apex and eave_end vertices
+    #--------------------------------------------------------------------------------
+    if not isinstance(gest_seg_np, np.ndarray):
+        gest_seg_np = np.array(gest_seg_np)
+    vertices = []
+    # Apex
+    apex_color = np.array(gestalt_color_mapping['apex'])
+    apex_mask = cv2.inRange(gest_seg_np, apex_color-0.5, apex_color+0.5)
+    if apex_mask.sum() > 0:
+        output = cv2.connectedComponentsWithStats(apex_mask, 8, cv2.CV_32S)
+        (numLabels, labels, stats, centroids) = output
+        stats, centroids = stats[1:], centroids[1:]  # skip background
+        for i in range(numLabels-1):
+            vert = {"xy": centroids[i], "type": "apex"}
+            vertices.append(vert)
+
+    # Eave end
+    eave_end_color = np.array(gestalt_color_mapping['eave_end_point'])
+    eave_end_mask = cv2.inRange(gest_seg_np, eave_end_color-0.5, eave_end_color+0.5)
+    if eave_end_mask.sum() > 0:
+        output = cv2.connectedComponentsWithStats(eave_end_mask, 8, cv2.CV_32S)
+        (numLabels, labels, stats, centroids) = output
+        stats, centroids = stats[1:], centroids[1:]
+        for i in range(numLabels-1):
+            vert = {"xy": centroids[i], "type": "eave_end_point"}
+            vertices.append(vert)
+
+    flashing_end_color = np.array(gestalt_color_mapping['flashing_end_point'])
+    flashing_end_mask = cv2.inRange(gest_seg_np, flashing_end_color - 0.5, flashing_end_color + 0.5)
+    if flashing_end_mask.sum() > 0:
+        output = cv2.connectedComponentsWithStats(flashing_end_mask, 8, cv2.CV_32S)
+        (numLabels, labels, stats, centroids) = output
+        if numLabels > 1:
+            stats_fl, centroids_fl = stats[1:], centroids[1:]
+            for i in range(numLabels - 1):
+                vert = {"xy": centroids_fl[i], "type": "flashing_end_point"}
+                vertices.append(vert)
+
+    # Consolidate apex points as array:
+    apex_pts = []
+    apex_idx_map = []  # keep track of index in 'vertices'
+    for idx, v in enumerate(vertices):
+        apex_pts.append(v['xy'])
+        apex_idx_map.append(idx)
+    apex_pts = np.array(apex_pts)
+
+    connections = []
+    edge_classes = ['eave', 'ridge', 'rake', 'valley', 'flashing', 'hip', 'step_flashing', 'transition_line']
+    for edge_class in edge_classes:
+        edge_color = np.array(gestalt_color_mapping[edge_class])
+        mask_raw = cv2.inRange(gest_seg_np, edge_color-0.5, edge_color+0.5)
+        # Possibly do morphological open/close to avoid merges or small holes
+        # open/close makes reusults worse 
+        kernel = np.ones((5, 5), np.uint8)  # smaller kernel to reduce over-merge
+        mask = cv2.morphologyEx(mask_raw, cv2.MORPH_CLOSE, kernel)
+        if mask.sum() == 0:
+            continue
+
+        # Connected components
+        output = cv2.connectedComponentsWithStats(mask, 8, cv2.CV_32S)
+        (numLabels, labels, stats, centroids) = output
+        # skip the background
+        stats, centroids = stats[1:], centroids[1:]
+        label_indices = range(1, numLabels)
+
+        # For each connected component, do a line fit
+        for lbl in label_indices:
+            ys, xs = np.where(labels == lbl)
+            if len(xs) < 2:
+                continue
+            # Fit a line using cv2.fitLine
+            pts_for_fit = np.column_stack([xs, ys]).astype(np.float32)
+            # (vx, vy, x0, y0) = direction + a point on the line
+            line_params = cv2.fitLine(pts_for_fit, distType=cv2.DIST_L2, 
+                                      param=0, reps=0.01, aeps=0.01)
+            vx, vy, x0, y0 = line_params.ravel()
+            # We'll approximate endpoints by projecting (xs, ys) onto the line,
+            # then taking min and max in the 1D param along the line.
+
+            # param along the line = ( (x - x0)*vx + (y - y0)*vy )
+            proj = ( (xs - x0)*vx + (ys - y0)*vy )
+            proj_min, proj_max = proj.min(), proj.max()
+            p1 = np.array([x0 + proj_min*vx, y0 + proj_min*vy])
+            p2 = np.array([x0 + proj_max*vx, y0 + proj_max*vy])
+
+            #--------------------------------------------------------------------------------
+            # Step C: If apex points are within 'edge_th' of segment, they are connected
+            #--------------------------------------------------------------------------------
+            if len(apex_pts) < 2:
+                continue
+
+            # Distance from each apex to the line segment
+            dists = np.array([
+                point_to_segment_dist(apex_pts[i], p1, p2)
+                for i in range(len(apex_pts))
+            ])
+
+            # Indices of apex points that are near
+            near_mask = (dists <= edge_th)
+            near_indices = np.where(near_mask)[0]
+            if len(near_indices) < 2:
+                continue
+
+            # Connect each pair among these near apex points
+            for i in range(len(near_indices)):
+                for j in range(i+1, len(near_indices)):
+                    a_idx = near_indices[i]
+                    b_idx = near_indices[j]
+                    # 'a_idx' and 'b_idx' are indices in apex_pts / apex_idx_map
+                    vA = apex_idx_map[a_idx]
+                    vB = apex_idx_map[b_idx]
+                    # Store the connection using sorted indexing
+                    conn = tuple(sorted((vA, vB)))
+                    connections.append(conn)
+
+    return vertices, connections
+
+
+def get_uv_depth(vertices: List[dict],
+                 depth_fitted: np.ndarray,
+                 sparse_depth: np.ndarray,
+                 search_radius: int = 10) -> Tuple[np.ndarray, np.ndarray]:
+    """
+    For each vertex, returns a 2D array of (u,v) and a matching 1D array of depths.
+    
+    We attempt to use the sparse_depth if available in a local neighborhood:
+      1. For each vertex coordinate (x, y), define a local window in sparse_depth 
+         of size (2*search_radius + 1).
+      2. Collect all valid (nonzero) values in that window.
+      3. If any exist, we take the *closest* valid pixel's depth.
+      4. Otherwise, we use depth_fitted[y, x].
+    
+    Parameters
+    ----------
+    vertices : List[dict]
+        Each dict must have "xy" at least, e.g. {"xy": (x, y), ...}
+    depth_fitted : np.ndarray
+        A 2D array (H, W), the dense (or corrected) depth for fallback.
+    sparse_depth : np.ndarray
+        A 2D array (H, W), mostly zeros except where accurate data is available.
+    search_radius : int
+        Pixel radius around the vertex in which to look for sparse depth values.
+    
+    Returns
+    -------
+    uv : np.ndarray of shape (N, 2)
+        2D float coordinates of each vertex (x, y).
+    vertex_depth : np.ndarray of shape (N,)
+        Depth value chosen for each vertex.
+    """
+    
+    # Collect each vertex's (x, y)
+    uv = np.array([vert['xy'] for vert in vertices], dtype=np.float32)
+    
+    # Convert to integer pixel coordinates (round or floor)
+    uv_int = np.round(uv).astype(np.int32)
+    H, W = depth_fitted.shape[:2]
+    
+    # Clip coordinates to stay within image bounds
+    uv_int[:, 0] = np.clip(uv_int[:, 0], 0, W - 1)
+    uv_int[:, 1] = np.clip(uv_int[:, 1], 0, H - 1)
+    
+    # Prepare output array of depths
+    vertex_depth = np.zeros(len(vertices), dtype=np.float32)
+    dense_count = 0
+    
+    for i, (x_i, y_i) in enumerate(uv_int):
+        # Local region in [x_i - search_radius, x_i + search_radius]
+        x0 = max(0, x_i - search_radius)
+        x1 = min(W, x_i + search_radius + 1)
+        y0 = max(0, y_i - search_radius)
+        y1 = min(H, y_i + search_radius + 1)
+        
+        # Crop out the local window in sparse_depth
+        region = sparse_depth[y0:y1, x0:x1]
+        
+        # Find all valid (non-zero) depths
+        valid_mask = (region > 0)
+        valid_y, valid_x = np.where(valid_mask)
+        
+        if valid_y.size > 0:
+            # Compute global coordinates for each valid pixel
+            global_x = x0 + valid_x
+            global_y = y0 + valid_y
+            
+            # Compute squared distance to center (x_i, y_i)
+            dist_sq = (global_x - x_i)**2 + (global_y - y_i)**2
+            
+            # Find the nearest valid pixel
+            min_idx = np.argmin(dist_sq)
+            nearest_depth = region[valid_y[min_idx], valid_x[min_idx]]
+            vertex_depth[i] = nearest_depth
+        else:
+            # Fallback to the dense depth
+            vertex_depth[i] = depth_fitted[y_i, x_i]
+            dense_count += 1
+    return uv, vertex_depth
+
+
+
+def project_vertices_to_3d(uv: np.ndarray, depth_vert: np.ndarray, col_img: pycolmap.Image) -> np.ndarray:
+    """
+    Projects 2D vertex coordinates with associated depths to 3D world coordinates.
+
+    Parameters
+    ----------
+    uv : np.ndarray
+        (N, 2) array of 2D vertex coordinates (u, v).
+    depth_vert : np.ndarray
+        (N,) array of depth values for each vertex.
+    col_img : pycolmap.Image
+
+    Returns
+    -------
+    vertices_3d : np.ndarray
+        (N, 3) array of vertex coordinates in 3D world space.
+    """
+    # Backproject to 3D local camera coordinates
+    xy_local = np.ones((len(uv), 3))
+    K = col_img.camera.calibration_matrix()
+    xy_local[:, 0] = (uv[:, 0] - K[0, 2]) / K[0, 0]
+    xy_local[:, 1] = (uv[:, 1] - K[1, 2]) / K[1, 1]
+    # Get the 3D vertices
+    vertices_3d_local = xy_local * depth_vert[...,None]
+    
+    # Create camera-to-world transformation matrix
+    world_to_cam = np.eye(4)
+    world_to_cam[:3] = col_img.cam_from_world.matrix()
+    cam_to_world = np.linalg.inv(world_to_cam)
+    
+    # Transform local 3D points to world coordinates
+    vertices_3d_homogeneous = cv2.convertPointsToHomogeneous(vertices_3d_local)
+    vertices_3d = cv2.transform(vertices_3d_homogeneous, cam_to_world)
+    vertices_3d = cv2.convertPointsFromHomogeneous(vertices_3d).reshape(-1, 3)
+    return vertices_3d
+
+
+def create_3d_wireframe_single_image(vertices: List[dict],
+                                     connections: List[Tuple[int, int]],
+                                     depth: PImage,
+                                     colmap_rec: pycolmap.Reconstruction,
+                                     img_id: str,
+                                     ade_seg: PImage) -> np.ndarray:
+    """
+    Processes a single image view to generate 3D vertex coordinates from existing 2D vertices/edges.
+
+    Parameters
+    ----------
+    vertices : List[dict]
+        List of 2D vertex dictionaries (e.g., {"xy": (x, y), "type": ...}).
+    connections : List[Tuple[int, int]]
+        List of 2D edge connections (indices into the vertices list).
+    depth : PIL.Image
+        Initial dense depth map as a PIL Image.
+    colmap_rec : pycolmap.Reconstruction
+        COLMAP reconstruction data.
+    img_id : str
+        Identifier for the current image within the COLMAP reconstruction.
+    ade_seg : PIL.Image
+        ADE20k segmentation map for the image.
+
+    Returns
+    -------
+    vertices_3d : np.ndarray
+        (N, 3) array of vertex coordinates in 3D world space.
+        Returns an empty array if processing fails (e.g., missing sparse depth).
+    """
+    # Check if initial vertices/connections are valid
+    if (len(vertices) < 2) or (len(connections) < 1):
+        # This case should ideally be handled before calling, but good to double check.
+        print(f'Warning: create_3d_wireframe_single_image called with insufficient vertices/connections for image {img_id}')
+        return np.empty((0, 3))
+
+    # Get fitted dense depth and sparse depth
+    depth_fitted, depth_sparse, found_sparse, col_img = get_fitted_dense_depth(
+        depth, colmap_rec, img_id, ade_seg
+    )
+
+    # Get UV coordinates and depth for each vertex
+    uv, depth_vert = get_uv_depth(vertices, depth_fitted, depth_sparse, search_radius=15) # default=10, 15*
+
+    # Backproject to 3D
+    vertices_3d = project_vertices_to_3d(uv, depth_vert, col_img)
+
+    return vertices_3d
+
+
+def merge_vertices_3d(vert_edge_per_image, th=0.5):
+    '''Merge vertices that are close to each other in 3D space and are of same types'''
+    # Initialize structures to collect vertices and connections from all images
+    all_3d_vertices = []
+    connections_3d = []
+    all_indexes = []
+    cur_start = 0
+    types = []
+    
+    # Combine vertices and update connection indices across all images
+    for cimg_idx, (vertices, connections, vertices_3d) in vert_edge_per_image.items():
+        types += [int(v['type']=='apex') for v in vertices]
+        all_3d_vertices.append(vertices_3d)
+        connections_3d+=[(x+cur_start,y+cur_start) for (x,y) in connections]
+        cur_start+=len(vertices_3d)
+    all_3d_vertices = np.concatenate(all_3d_vertices, axis=0)
+    
+    # Calculate distance matrix between all vertices
+    distmat = cdist(all_3d_vertices, all_3d_vertices)
+    types = np.array(types).reshape(-1,1)
+    same_types = cdist(types, types)
+    
+    # Create mask for vertices that should be merged (close in space and same type)
+    mask_to_merge = (distmat <= th) & (same_types==0)
+    new_vertices = []
+    new_connections = []
+    
+    # Extract vertex indices to merge based on the mask
+    to_merge = sorted(list(set([tuple(a.nonzero()[0].tolist()) for a in mask_to_merge])))
+    
+    # Build groups of vertices to merge (transitive grouping)
+    to_merge_final = defaultdict(list)
+    for i in range(len(all_3d_vertices)):
+        for j in to_merge:
+            if i in j:
+                to_merge_final[i]+=j
+    
+    # Remove duplicates in each group
+    for k, v in to_merge_final.items():
+        to_merge_final[k] = list(set(v))
+    
+    # Create final merge groups without duplicates
+    already_there = set() 
+    merged = []
+    for k, v in to_merge_final.items():
+        if k in already_there:
+            continue
+        merged.append(v)
+        for vv in v:
+            already_there.add(vv)
+    
+    # Calculate new vertex positions (average of merged groups)
+    old_idx_to_new = {}
+    count=0
+    for idxs in merged:
+        new_vertices.append(all_3d_vertices[idxs].mean(axis=0))
+        for idx in idxs:
+            old_idx_to_new[idx] = count
+        count +=1
+    new_vertices=np.array(new_vertices)
+    
+    # Update connections to use new vertex indices
+    for conn in connections_3d:
+        new_con = sorted((old_idx_to_new[conn[0]], old_idx_to_new[conn[1]]))
+        if new_con[0] == new_con[1]:
+            continue
+        if new_con not in new_connections:
+            new_connections.append(new_con)
+    return new_vertices, new_connections
+
+
+def prune_not_connected(all_3d_vertices, connections_3d, keep_largest=True):
+    """
+    Prune vertices not connected to anything. If keep_largest=True, also
+    keep only the largest connected component in the graph.
+    """
+    if len(all_3d_vertices) == 0:
+        return np.array([]), []
+
+    # adjacency
+    adj = defaultdict(set)
+    for (i, j) in connections_3d:
+        adj[i].add(j)
+        adj[j].add(i)
+
+    # keep only vertices that appear in at least one edge
+    used_idxs = set()
+    for (i, j) in connections_3d:
+        used_idxs.add(i)
+        used_idxs.add(j)
+
+    if not used_idxs:
+        return np.empty((0,3)), []
+
+    # If we only want to remove truly isolated points, but keep multiple subgraphs:
+    if not keep_largest:
+        new_map = {}
+        used_list = sorted(list(used_idxs))
+        for new_id, old_id in enumerate(used_list):
+            new_map[old_id] = new_id
+        new_vertices = np.array([all_3d_vertices[old_id] for old_id in used_list])
+        new_conns = []
+        for (i, j) in connections_3d:
+            if i in used_idxs and j in used_idxs:
+                new_conns.append((new_map[i], new_map[j]))
+        return new_vertices, new_conns
+
+    # Otherwise find the largest connected component:
+    visited = set()
+    def bfs(start):
+        queue = [start]
+        comp = []
+        visited.add(start)
+        while queue:
+            cur = queue.pop()
+            comp.append(cur)
+            for neigh in adj[cur]:
+                if neigh not in visited:
+                    visited.add(neigh)
+                    queue.append(neigh)
+        return comp
+
+    # Collect all subgraphs
+    comps = []
+    for idx in used_idxs:
+        if idx not in visited:
+            c = bfs(idx)
+            comps.append(c)
+
+    # pick largest
+    comps.sort(key=lambda c: len(c), reverse=True)
+    largest = comps[0] if len(comps)>0 else []
+
+    # Remap
+    new_map = {}
+    for new_id, old_id in enumerate(largest):
+        new_map[old_id] = new_id
+
+    new_vertices = np.array([all_3d_vertices[old_id] for old_id in largest])
+    new_conns = []
+    for (i, j) in connections_3d:
+        if i in largest and j in largest:
+            new_conns.append((new_map[i], new_map[j]))
+
+    # remove duplicates
+    new_conns = list(set([tuple(sorted(c)) for c in new_conns]))
+    return new_vertices, new_conns
+
+def get_sparse_depth(colmap_rec, img_id_substring, depth):
+    """
+    Return a sparse depth map for the COLMAP image whose name contains
+    `img_id_substring`. The output is an array of shape `depth_shape` (H,W),
+    where only the projected 3D points get a depth > 0, else 0.
+    """
+    H, W = depth.shape
+
+    # 1) Find the matching COLMAP image
+    found_img = None
+    for img_id_c, col_img in colmap_rec.images.items():
+        if img_id_substring in col_img.name:
+            found_img = col_img
+            break
+    if found_img is None:
+        print(f"Image substring {img_id_substring} not found in COLMAP.")
+        return np.zeros((H, W), dtype=np.float32), False, None
+    
+    # 2) Gather 3D points that this image sees
+    points_xyz = []
+    for pid, p3D in colmap_rec.points3D.items():
+        if found_img.has_point3D(pid):
+            points_xyz.append(p3D.xyz)  # world coords
+    if not points_xyz:
+        print(f"No 3D points associated with {found_img.name}.")
+        return np.zeros((H, W), dtype=np.float32), False, found_img
+    
+    points_xyz = np.array(points_xyz)  # (N, 3)
+    
+    # 3) For each point, project via col_img.project_point()
+    uv = []
+    z_vals = []
+    for xyz in points_xyz:
+        proj = found_img.project_point(xyz)  # returns (u, v) in image coords or None
+        if proj is not None:
+            u_i, v_i = proj
+            u_i = int(round(u_i))
+            v_i = int(round(v_i))
+            # Check in-bounds
+            if 0 <= u_i < W and 0 <= v_i < H:
+                uv.append((u_i, v_i))
+                # We'll compute depth as Z in camera coords
+                # from the world->cam transform col_img holds
+                mat4x4 = np.eye(4)
+                mat4x4[:3, :4] = found_img.cam_from_world.matrix()
+                p_cam =  mat4x4@ np.array([xyz[0], xyz[1], xyz[2], 1.0])
+                z_vals.append(p_cam[2] / p_cam[3]) 
+    
+    uv = np.array(uv, dtype=int)     # shape (M,2)
+    z_vals = np.array(z_vals)        # shape (M,)
+    
+    depth_out = np.zeros((H, W), dtype=np.float32)
+    depth_out[uv[:,1], uv[:,0]] = z_vals  # Note: uv = (u, v), so row = v, col = u
+    
+    return depth_out, True, found_img
+
+
+def fit_scale_robust_median(depth, sparse_depth, validity_mask=None):
+    """
+    Fit a scale factor to the depth map using the median of the ratio of sparse to dense depth.
+    """
+    if validity_mask is None:
+        mask = (sparse_depth != 0)
+    else:
+        mask = (sparse_depth != 0) & validity_mask
+    mask = mask & (depth <50) & (sparse_depth <50)
+    X = depth[mask]
+    Y = sparse_depth[mask]
+    alpha =np.median(Y/X)
+    depth_fitted = alpha * depth
+    return alpha, depth_fitted
+    
+
+def get_fitted_dense_depth(depth, colmap_rec, img_id, ade20k_seg):
+    """
+    Gets sparse depth from COLMAP, computes a house mask, fits dense depth to sparse 
+    depth within the mask, and returns the fitted dense depth.
+
+    Parameters
+    ----------
+    depth : np.ndarray
+        Initial dense depth map (H, W).
+    colmap_rec : pycolmap.Reconstruction
+        COLMAP reconstruction data.
+    img_id : str
+        Identifier for the current image within the COLMAP reconstruction.
+    K : np.ndarray
+        Camera intrinsic matrix (3x3).
+    R : np.ndarray
+        Camera rotation matrix (3x3).
+    t : np.ndarray
+        Camera translation vector (3,).
+    ade20k_seg : PIL.Image
+        ADE20k segmentation map for the image.
+
+    Returns
+    -------
+    depth_fitted : np.ndarray
+        Dense depth map scaled and shifted to align with sparse depth within the house mask (H, W).
+    depth_sparse : np.ndarray
+        The sparse depth map obtained from COLMAP (H, W).
+    found_sparse : bool
+        True if sparse depth points were found for this image, False otherwise.
+    """
+    depth_np = np.array(depth) / 1000. # Convert mm to meters if needed
+    depth_sparse, found_sparse, col_img = get_sparse_depth(colmap_rec, img_id, depth_np)
+    
+    if not found_sparse:
+        print(f'No sparse depth found for image {img_id}')
+        # Return original (meter-scaled) depth if no sparse data
+        return depth_np, np.zeros_like(depth_np), False, None
+
+    # Get house mask to focus fitting on relevant areas
+    house_mask = get_house_mask(ade20k_seg)
+    
+    # Fit dense depth to sparse depth (scale only), using only points within the house mask
+    k, depth_fitted = fit_scale_robust_median(depth_np, depth_sparse, validity_mask=house_mask)
+    print(f"Fitted depth scale k={k:.4f} for image {img_id}")
+    #depth_fitted = depth_np# * house_mask.astype(np.float32)
+    depth_sparse = depth_sparse# * house_mask.astype(np.float32)
+    return depth_fitted, depth_sparse, True, col_img
+
+
+def prune_too_far(all_3d_vertices, connections_3d, colmap_rec, th = 3.0):
+    """
+    Prune vertices that are too far from sparse point cloud
+    
+    """
+    xyz_sfm=[]
+    for k, v in colmap_rec.points3D.items():
+        xyz_sfm.append(v.xyz)
+    xyz_sfm = np.array(xyz_sfm)
+    distmat = cdist(all_3d_vertices, xyz_sfm)
+    mindist = distmat.min(axis=1)
+    mask = mindist <= th
+    all_3d_vertices_new = all_3d_vertices[mask]
+    old_idx_survived = np.arange(len(all_3d_vertices))[mask]
+    new_idxs = np.arange(len(all_3d_vertices_new))
+    old_to_new_idx = dict(zip(old_idx_survived, new_idxs))
+    connections_3d_new = [(old_to_new_idx[conn[0]], old_to_new_idx[conn[1]]) for conn in connections_3d if mask[conn[0]] and mask[conn[1]]]   
+    return all_3d_vertices_new, connections_3d_new
+
+
+def predict_wireframe(entry) -> Tuple[np.ndarray, List[int]]:
+    """
+    Predict 3D wireframe from a dataset entry.
+    """
+    good_entry = convert_entry_to_human_readable(entry)
+    vert_edge_per_image = {}
+    for i, (gest, depth, K, R, t, img_id, ade_seg) in enumerate(zip(good_entry['gestalt'],
+                                                good_entry['depth'], 
+                                                good_entry['K'],
+                                                good_entry['R'],
+                                                good_entry['t'],
+                                                good_entry['image_ids'],
+                                                good_entry['ade'] # Added ade20k segmentation
+                                                )):
+        colmap_rec = good_entry['colmap_binary']
+        K = np.array(K)
+        R = np.array(R)
+        t = np.array(t)
+        # Resize gestalt segmentation to match depth map size
+        depth_size = (np.array(depth).shape[1], np.array(depth).shape[0]) # W, H
+        gest_seg = gest.resize(depth_size)
+        gest_seg_np = np.array(gest_seg).astype(np.uint8)
+        
+        # Get 2D vertices and edges first
+        vertices, connections = get_vertices_and_edges_from_segmentation(gest_seg_np, edge_th=15.) # default 10, 15*
+        
+        # Check if we have enough to proceed
+        if (len(vertices) < 2) or (len(connections) < 1):
+            print(f'Not enough vertices or connections found in image {i}, skipping.')
+            vert_edge_per_image[i] = [], [], np.empty((0, 3))
+            continue
+            
+        # Call the refactored function to get 3D points
+        vertices_3d = create_3d_wireframe_single_image(
+            vertices, connections, depth, colmap_rec, img_id, ade_seg
+        )
+        # Store original 2D vertices, connections, and computed 3D points
+        vert_edge_per_image[i] = vertices, connections, vertices_3d
+    
+    # Merge vertices from all images
+        all_3d_vertices, connections_3d = merge_vertices_3d(vert_edge_per_image, th=0.4) # default=0.5, 0.4*
+                                                                                     # tighten the 3D merge radius
+
+    all_3d_vertices_clean, connections_3d_clean  = prune_not_connected(all_3d_vertices, connections_3d, keep_largest=False)
+    all_3d_vertices_clean, connections_3d_clean  = prune_too_far(all_3d_vertices_clean, connections_3d_clean, colmap_rec, th = 4.0) # default=4.0*
+    
+    if (len(all_3d_vertices_clean) < 2) or len(connections_3d_clean) < 1:
+        print (f'Not enough vertices or connections in the 3D vertices')
+        return empty_solution()
+
+    return all_3d_vertices_clean, connections_3d_clean 

tools2025/hoho2025/hoho.py

@@ -0,0 +1,340 @@
+import os 
+import json
+import shutil
+from pathlib import Path
+from typing import Dict
+import warnings 
+import contextlib
+import tempfile 
+from PIL import Image
+import io
+import webdataset as wds 
+import numpy as np
+import importlib
+import subprocess
+
+
+from PIL import ImageFile
+
+from huggingface_hub.utils._headers import build_hf_headers # note: using _headers
+
+
+ImageFile.LOAD_TRUNCATED_IMAGES = True
+
+LOCAL_DATADIR = None
+
+def setup(local_dir='./data/usm-training-data/data'):
+    
+    # If we are in the test environment, we need to link the data directory to the correct location
+    tmp_datadir = Path('/tmp/data/data')
+    local_test_datadir = Path('./data/usm-test-data-x/data')
+    local_val_datadir = Path(local_dir)
+    
+    os.system('pwd')
+    os.system('ls -lahtr .')
+    
+    if tmp_datadir.exists() and not local_test_datadir.exists():
+        global LOCAL_DATADIR
+        LOCAL_DATADIR = local_test_datadir
+        # shutil.move(datadir, './usm-test-data-x/data')
+        print(f"Linking {tmp_datadir} to {LOCAL_DATADIR} (we are in the test environment)")
+        LOCAL_DATADIR.parent.mkdir(parents=True, exist_ok=True)
+        LOCAL_DATADIR.symlink_to(tmp_datadir)
+    else:
+        LOCAL_DATADIR = local_val_datadir
+        print(f"Using {LOCAL_DATADIR} as the data directory (we are running locally)")
+    
+    if not LOCAL_DATADIR.exists():
+        warnings.warn(f"Data directory {LOCAL_DATADIR} does not exist: creating it...")
+        LOCAL_DATADIR.mkdir(parents=True)
+    
+    return LOCAL_DATADIR
+
+
+def download_package(package_name, path_to_save='packages'):
+    """
+    Downloads a package using pip and saves it to a specified directory.
+
+    Parameters:
+    package_name (str): The name of the package to download.
+    path_to_save (str): The path to the directory where the package will be saved.
+    """
+    try:
+        # pip download webdataset -d packages/webdataset --platform manylinux1_x86_64 --python-version 38 --only-binary=:all:
+        subprocess.check_call([subprocess.sys.executable, "-m", "pip", "download", package_name, 
+                               "-d", str(Path(path_to_save)/package_name),  # Download the package to the specified directory
+                               "--platform", "manylinux1_x86_64",  # Specify the platform
+                               "--python-version", "38",  # Specify the Python version
+                               "--only-binary=:all:"])  # Download only binary packages
+        print(f'Package "{package_name}" downloaded successfully')
+    except subprocess.CalledProcessError as e:
+        print(f'Failed to downloaded package "{package_name}". Error: {e}')
+              
+              
+def install_package_from_local_file(package_name, folder='packages'):
+    """
+    Installs a package from a local .whl file or a directory containing .whl files using pip.
+
+    Parameters:
+    path_to_file_or_directory (str): The path to the .whl file or the directory containing .whl files.
+    """
+    try:
+        pth = str(Path(folder) / package_name)
+        subprocess.check_call([subprocess.sys.executable, "-m", "pip", "install", 
+                               "--no-index",  # Do not use package index
+                               "--find-links", pth,  # Look for packages in the specified directory or at the file
+                               package_name])  # Specify the package to install
+        print(f"Package installed successfully from {pth}")
+    except subprocess.CalledProcessError as e:
+        print(f"Failed to install package from {pth}. Error: {e}")
+        
+        
+def importt(module_name, as_name=None):
+    """
+    Imports a module and returns it.
+
+    Parameters:
+    module_name (str): The name of the module to import.
+    as_name (str): The name to use for the imported module. If None, the original module name will be used.
+
+    Returns:
+    The imported module.
+    """
+    for _ in range(2):
+        try:
+            if as_name is None:
+                print(f'imported {module_name}')
+                return importlib.import_module(module_name)
+            else:
+                print(f'imported {module_name} as {as_name}')
+                return importlib.import_module(module_name, as_name)
+        except ModuleNotFoundError as e:
+            install_package_from_local_file(module_name)
+            print(f"Failed to import module {module_name}. Error: {e}")
+            
+    
+def prepare_submission():
+    # Download packages from requirements.txt 
+    if Path('requirements.txt').exists():
+        print('downloading packages from requirements.txt')
+        Path('packages').mkdir(exist_ok=True)
+        with open('requirements.txt') as f:
+            packages = f.readlines()
+            for p in packages:
+                download_package(p.strip())
+                
+    print('all packages downloaded. Don\'t foget to include the packages in the submission by adding them with git lfs.')
+        
+
+def Rt_to_eye_target(im, K, R, t):
+    height = im.height
+    focal_length = K[0,0]
+    fov = 2.0 * np.arctan2((0.5 * height), focal_length) / (np.pi / 180.0)
+
+    x_axis, y_axis, z_axis = R
+
+    eye = -(R.T @ t).squeeze()
+    z_axis = z_axis.squeeze()
+    target = eye + z_axis
+    up = -y_axis
+    
+    return eye, target, up, fov        
+
+
+########## general utilities ##########
+
+
+@contextlib.contextmanager
+def working_directory(path):
+    """Changes working directory and returns to previous on exit."""
+    prev_cwd = Path.cwd()
+    os.chdir(path)
+    try:
+        yield
+    finally:
+        os.chdir(prev_cwd)
+        
+@contextlib.contextmanager
+def temp_working_directory():
+    with tempfile.TemporaryDirectory(dir='.') as D:
+        with working_directory(D):
+            yield
+
+
+############# Dataset ############# 
+def proc(row, split='train'):
+    out = {}
+    out['__key__'] = None
+    out['__imagekey__'] = []
+    for k, v in row.items():
+        key_parts = k.split('.')
+        colname = key_parts[0]
+        if colname == 'ade20k': 
+            out['__imagekey__'].append(key_parts[1])
+        if colname in {'ade20k', 'depthcm', 'gestalt'}:
+            if colname in out:
+                out[colname].append(v)
+            else:
+                out[colname] = [v]
+        elif colname in {'wireframe', 'mesh'}:
+            out.update({a: b for a,b in v.items()})
+        elif colname in 'kr':
+            out[colname.upper()] = v
+        else:
+            out[colname] = v
+    return Sample(out)
+
+
+
+def decode_colmap(s):
+    import hoho2025.read_write_colmap as read_write_colmap
+    with temp_working_directory():
+
+        with open('points3D.bin', 'wb') as stream:
+            stream.write(s['points3d'])
+
+
+        with open('cameras.bin', 'wb') as stream:
+            stream.write(s['cameras'])
+
+
+        with open('images.bin', 'wb') as stream:
+            stream.write(s['images'])
+    
+        
+        cameras, images, points3D = read_write_colmap.read_model(
+            path='.', ext='.bin'
+        )
+    return cameras, images, points3D
+
+
+def decode(row):
+    cameras, images, points3D = decode_colmap(row)
+    
+    out = {}
+    
+    for k, v in row.items():
+        # colname = k.split('.')[0]
+        if k in {'ade20k', 'depthcm', 'gestalt'}:
+            # print(k, len(v), type(v))
+            v = [Image.open(io.BytesIO(im)) for im in v]
+            if k in out:
+                out[k].extend(v)
+            else:
+                out[k] = v
+        elif k in {'wireframe', 'mesh'}:
+            # out.update({a: b.tolist() for a,b in v.items()})
+            v = dict(np.load(io.BytesIO(v)))
+            out.update({a: b for a,b in v.items()})
+        elif k in 'kr':
+            out[k.upper()] = v
+        elif k == 'cameras':
+            out[k] = cameras
+        elif k == 'images':
+            out[k] = images
+        elif k =='points3d':
+            out[k] = points3D
+        else:
+            out[k] = v
+            
+    return Sample(out)
+
+
+class Sample(Dict):
+    def __repr__(self):
+        return str({k: v.shape if hasattr(v, 'shape') else [type(v[0])] if isinstance(v, list) else type(v) for k,v in self.items()})
+
+    
+        
+def get_params():
+    exmaple_param_dict = {
+        "competition_id": "usm3d/S23DR",
+        "competition_type": "script",
+        "metric": "custom",
+        "token": "hf_**********************************",
+        "team_id": "local-test-team_id",
+        "submission_id": "local-test-submission_id",
+        "submission_id_col": "__key__",
+        "submission_cols": [
+            "__key__",
+            "wf_edges",
+            "wf_vertices",
+            "edge_semantics"
+        ],
+        "submission_rows": 180,
+        "output_path": ".",
+        "submission_repo": "<THE HF MODEL ID of THIS REPO",
+        "time_limit": 7200,
+        "dataset": "usm3d/usm-test-data-x",
+        "submission_filenames": [
+            "submission.parquet"
+        ]
+    }
+    
+    param_path = Path('params.json')
+    
+    if not param_path.exists():
+        print('params.json not found (this means we probably aren\'t in the test env). Using example params.')
+        params = exmaple_param_dict
+    else:
+        print('found params.json (this means we are probably in the test env). Using params from file.')
+        with param_path.open() as f:
+            params = json.load(f)
+    print(params)
+    return params
+
+
+
+
+
+SHARD_IDS = {'train': (0, 25), 'val': (25, 26), 'public': (26, 27), 'private': (27, 32)}
+def get_dataset(decode='pil', proc=proc, split='train', dataset_type='webdataset', stream=True):
+    if LOCAL_DATADIR is None:
+        raise ValueError('LOCAL_DATADIR is not set. Please run setup() first.')
+        
+    local_dir = Path(LOCAL_DATADIR)
+    if split != 'all':
+        local_dir = local_dir / split
+    
+    paths = [str(p) for p in local_dir.rglob('*.tar.gz')]
+    msg = f'no tarfiles found in {local_dir}.'
+    if len(paths) == 0:
+        if stream:
+            if split=='all': split = 'train'
+            warnings.warn('streaming isn\'t using with \'all\': changing `split` to \'train\'')
+            warnings.warn(msg)
+            if split == 'val':
+                names = [f'data/val/inputs/hoho_v3_{i:03}-of-032.tar.gz' for i in range(*SHARD_IDS[split])]
+            elif split == 'train':
+                names = [f'data/train/hoho_v3_{i:03}-of-032.tar.gz' for i in range(*SHARD_IDS[split])]
+            
+            auth = build_hf_headers()['authorization']
+            paths = [f"pipe:curl -L -s https://huggingface.co/datasets/usm3d/hoho-train-set/resolve/main/{name} -H 'Authorization: {auth}'" for name in names]
+        else:
+            raise FileNotFoundError(msg)
+    
+    dataset = wds.WebDataset(paths)
+        
+    if decode is not None:
+        dataset = dataset.decode(decode)
+    else:
+        dataset = dataset.decode()
+    
+    dataset = dataset.map(proc)
+    
+    if dataset_type == 'webdataset':
+        return dataset
+    
+    if dataset_type == 'hf':
+        import datasets
+        from datasets import Features, Value, Sequence, Image, Array2D
+       
+        if split == 'train':
+            return datasets.IterableDataset.from_generator(lambda: dataset.iterator())
+        elif split == 'val':
+            return datasets.IterableDataset.from_generator(lambda: dataset.iterator())
+        else:
+            raise NotImplementedError('only train and val are implemented as hf datasets')
+
+    
+    

tools2025/hoho2025/metric_helper.py

@@ -0,0 +1,167 @@
+import numpy as np
+from scipy.spatial.distance import cdist
+from scipy.optimize import linear_sum_assignment
+import torch
+import trimesh
+from time import time
+
+MAX_SCORE = 1.0
+
+def get_one_primitive(p1, p2, c=(255, 0, 0), radius=25, primitive_type='cylinder', sections=6):
+    if len(c) == 1:
+        c = [c[0]] * 4
+    elif len(c) == 3:
+        c = [*c, 255]
+    elif len(c) != 4:
+        raise ValueError(f'{c} is not a valid color (must have 1,3, or 4 elements).')
+
+    p1, p2 = np.asarray(p1), np.asarray(p2)
+    l = np.linalg.norm(p2 - p1)
+    
+    # Add check for zero-length edges
+    if l < 1e-6:
+        return None
+        
+    direction = (p2 - p1) / l
+
+    T = np.eye(4)
+    T[:3, 2] = direction
+    T[:3, 3] = (p1 + p2) / 2
+
+    b0, b1 = T[:3, 0], T[:3, 1]
+    if np.abs(np.dot(b0, direction)) < np.abs(np.dot(b1, direction)):
+        T[:3, 1] = -np.cross(b0, direction)
+    else:
+        T[:3, 0] = np.cross(b1, direction)
+
+    if primitive_type == 'capsule':
+        mesh = trimesh.primitives.Capsule(radius=radius, height=l, transform=T, sections=sections)
+    elif primitive_type == 'cylinder':
+        mesh = trimesh.primitives.Cylinder(radius=radius, height=l, transform=T, sections=sections)
+    else:
+        raise ValueError("Unknown primitive!")
+
+    # Add vertex color initialization check
+    if not hasattr(mesh.visual, 'vertex_colors') or mesh.visual.vertex_colors is None:
+        mesh.visual.vertex_colors = np.ones((len(mesh.vertices), 4)) * 255
+        
+    mesh.visual.vertex_colors = np.ones_like(mesh.visual.vertex_colors) * c
+    return mesh
+
+def get_primitives(vertices, edges, radius=25, c=[255, 0, 0]):
+    # Convert vertices to a NumPy array
+    if isinstance(vertices, torch.Tensor):
+        vertices = vertices.detach().cpu().numpy()
+    else:
+        vertices = np.asarray(vertices)
+    
+    # Convert edges to a NumPy array of integers
+    if isinstance(edges, torch.Tensor):
+        edges = edges.detach().cpu().numpy().astype(np.int64)
+    else:
+        edges = np.asarray(edges, dtype=np.int64)
+    
+    primitives = []
+    for e in edges:
+        # Add edge validation
+        if e[0] >= len(vertices) or e[1] >= len(vertices):
+            continue
+        primitive = get_one_primitive(vertices[e[0]], vertices[e[1]], radius=radius, c=c)
+        if primitive is not None:
+            primitives.append(primitive)
+    return primitives
+
+
+
+def compute_mesh_iou_VOLUME(pd_vertices, pd_edges, gt_vertices, gt_edges, radius=20, engine='manifold'):
+    # check empty 
+    if len(pd_edges) == 0 or len(gt_edges) == 0:
+        return 0.0
+
+    pd_vertices = pd_vertices.detach().cpu() if isinstance(pd_vertices, torch.Tensor) else pd_vertices
+    pd_edges = pd_edges.detach().cpu() if isinstance(pd_edges, torch.Tensor) else pd_edges
+    gt_vertices = gt_vertices.detach().cpu() if isinstance(gt_vertices, torch.Tensor) else gt_vertices
+    gt_edges = gt_edges.detach().cpu() if isinstance(gt_edges, torch.Tensor) else gt_edges
+
+    pd_primitives = get_primitives(pd_vertices, pd_edges, radius=radius, c=[0, 255, 0])
+    gt_primitives = get_primitives(gt_vertices, gt_edges, radius=radius, c=[255, 0, 0])
+    # check for empty primitives
+    if not pd_primitives or not gt_primitives:
+        return 0.0
+
+    # Add bounding box check to detect non-overlapping cases quickly
+    pd_bounds = np.array([p.bounds for p in pd_primitives])
+    gt_bounds = np.array([p.bounds for p in gt_primitives])
+    
+    pd_min, pd_max = np.min(pd_bounds[:, 0], axis=0), np.max(pd_bounds[:, 1], axis=0)
+    gt_min, gt_max = np.min(gt_bounds[:, 0], axis=0), np.max(gt_bounds[:, 1], axis=0)
+    
+    # If bounding boxes don't overlap, return 0
+    if np.any(pd_max < gt_min) or np.any(pd_min > gt_max):
+        return 0.0
+    t=time()
+    mesh_pred = trimesh.boolean.union(pd_primitives, engine=engine)
+    #print(f"mesh_pred union: {time() - t} {mesh_pred.is_volume}")
+    t=time()
+    mesh_gt= trimesh.boolean.union(gt_primitives, engine=engine)
+    #print(f"mesh_gt union: {time() - t} {mesh_gt.is_volume}")
+
+    if mesh_pred.is_volume and mesh_gt.is_volume:
+        t=time()
+        inter_volume = trimesh.boolean.intersection([mesh_pred, mesh_gt], engine=engine).volume
+        #print(f"inter_volume: {time() - t}")
+    else:
+        all_inter = []
+        t=time()
+        for pd_prim in pd_primitives:
+            pd_min, pd_max = pd_prim.bounds
+            for gt_prim in gt_primitives:
+                # Skip intersection calculation if bounding boxes don't overlap
+                gt_min, gt_max = gt_prim.bounds
+                if np.any(pd_max < gt_min) or np.any(pd_min > gt_max):
+                    continue
+                inter = trimesh.boolean.intersection([pd_prim, gt_prim], engine=engine)
+                if inter.is_volume and inter.volume > 0:
+                    all_inter.append(inter)
+        inter_volume = trimesh.boolean.union(all_inter, engine=engine).volume if all_inter else 0
+        #print(f"all_inter: {time() - t}")
+    union_volume = mesh_pred.volume + mesh_gt.volume - inter_volume
+    
+    return inter_volume / union_volume if union_volume > 0 else 0.0
+
+
+# ----------------- Corner F1 -----------------
+def compute_ap_metrics(pd_vertices, gt_vertices, thresh=25):
+    if len(pd_vertices) == 0 or len(gt_vertices) == 0:
+        return 0.0
+
+    dists = cdist(pd_vertices, gt_vertices)
+    row_ind, col_ind = linear_sum_assignment(dists)
+
+    tp = (dists[row_ind, col_ind] <= thresh).sum()
+    precision = tp / len(pd_vertices) if len(pd_vertices) > 0 else 0
+    recall = tp / len(gt_vertices) if len(gt_vertices) > 0 else 0
+    denom = precision + recall
+    f1 = (2 * precision * recall / denom) if denom > 0 else 0.0
+    return f1
+
+def batch_corner_f1(X, Y, distance_thresh=25):
+    results = []
+    for (pd_v, _), (gt_v, _) in zip(X, Y):
+        results.append(compute_ap_metrics(pd_v, gt_v, thresh=distance_thresh))
+    return np.array(results)
+
+# ----------------- HSS Metric -----------------
+from collections import namedtuple
+HSSReturnType = namedtuple('HSSReturnType', ['hss', 'f1', 'iou'])
+def hss(y_hat_v, y_hat_e, y_v, y_e, vert_thresh=0.5, edge_thresh=0.5):
+    X = [(y_hat_v, y_hat_e)]
+    Y = [(y_v, y_e)]
+    t=time()
+    f1 = np.clip(batch_corner_f1(X, Y, distance_thresh=vert_thresh)[0], 0, 1)
+    #print(f"f1 {f1}: in {time() - t:.2f} sec")
+    t=time()
+    IoU = np.clip(compute_mesh_iou_VOLUME(y_hat_v, y_hat_e, y_v, y_e, radius=edge_thresh), 0, 1)
+    #print(f"IoU: {IoU} in {time() - t:.2f} sec")
+    score = 2 * f1 * IoU / (f1 + IoU) if (f1 + IoU) > 0 else 0.0
+    return HSSReturnType(hss=score, f1=f1, iou=IoU)

tools2025/hoho2025/read_write_colmap.py

@@ -0,0 +1,488 @@
+# Modified to read from bytes-like object by Dmytro Mishkin.
+# The original license is below:
+# Copyright (c) 2018, ETH Zurich and UNC Chapel Hill.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+#     * Redistributions of source code must retain the above copyright
+#       notice, this list of conditions and the following disclaimer.
+#
+#     * Redistributions in binary form must reproduce the above copyright
+#       notice, this list of conditions and the following disclaimer in the
+#       documentation and/or other materials provided with the distribution.
+#
+#     * Neither the name of ETH Zurich and UNC Chapel Hill nor the names of
+#       its contributors may be used to endorse or promote products derived
+#       from this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.
+#
+# Author: Johannes L. Schoenberger (jsch-at-demuc-dot-de)
+
+import os
+import collections
+import numpy as np
+import struct
+import argparse
+
+
+CameraModel = collections.namedtuple(
+    "CameraModel", ["model_id", "model_name", "num_params"])
+Camera = collections.namedtuple(
+    "Camera", ["id", "model", "width", "height", "params"])
+BaseImage = collections.namedtuple(
+    "Image", ["id", "qvec", "tvec", "camera_id", "name", "xys", "point3D_ids"])
+Point3D = collections.namedtuple(
+    "Point3D", ["id", "xyz", "rgb", "error", "image_ids", "point2D_idxs"])
+
+
+class Image(BaseImage):
+    def qvec2rotmat(self):
+        return qvec2rotmat(self.qvec)
+
+
+CAMERA_MODELS = {
+    CameraModel(model_id=0, model_name="SIMPLE_PINHOLE", num_params=3),
+    CameraModel(model_id=1, model_name="PINHOLE", num_params=4),
+    CameraModel(model_id=2, model_name="SIMPLE_RADIAL", num_params=4),
+    CameraModel(model_id=3, model_name="RADIAL", num_params=5),
+    CameraModel(model_id=4, model_name="OPENCV", num_params=8),
+    CameraModel(model_id=5, model_name="OPENCV_FISHEYE", num_params=8),
+    CameraModel(model_id=6, model_name="FULL_OPENCV", num_params=12),
+    CameraModel(model_id=7, model_name="FOV", num_params=5),
+    CameraModel(model_id=8, model_name="SIMPLE_RADIAL_FISHEYE", num_params=4),
+    CameraModel(model_id=9, model_name="RADIAL_FISHEYE", num_params=5),
+    CameraModel(model_id=10, model_name="THIN_PRISM_FISHEYE", num_params=12)
+}
+CAMERA_MODEL_IDS = dict([(camera_model.model_id, camera_model)
+                         for camera_model in CAMERA_MODELS])
+CAMERA_MODEL_NAMES = dict([(camera_model.model_name, camera_model)
+                           for camera_model in CAMERA_MODELS])
+
+
+def read_next_bytes(fid, num_bytes, format_char_sequence, endian_character="<"):
+    """Read and unpack the next bytes from a binary file.
+    :param fid:
+    :param num_bytes: Sum of combination of {2, 4, 8}, e.g. 2, 6, 16, 30, etc.
+    :param format_char_sequence: List of {c, e, f, d, h, H, i, I, l, L, q, Q}.
+    :param endian_character: Any of {@, =, <, >, !}
+    :return: Tuple of read and unpacked values.
+    """
+    data = fid.read(num_bytes)
+    return struct.unpack(endian_character + format_char_sequence, data)
+
+
+def write_next_bytes(fid, data, format_char_sequence, endian_character="<"):
+    """pack and write to a binary file.
+    :param fid:
+    :param data: data to send, if multiple elements are sent at the same time,
+    they should be encapsuled either in a list or a tuple
+    :param format_char_sequence: List of {c, e, f, d, h, H, i, I, l, L, q, Q}.
+    should be the same length as the data list or tuple
+    :param endian_character: Any of {@, =, <, >, !}
+    """
+    if isinstance(data, (list, tuple)):
+        bytes = struct.pack(endian_character + format_char_sequence, *data)
+    else:
+        bytes = struct.pack(endian_character + format_char_sequence, data)
+    fid.write(bytes)
+
+
+def read_cameras_text(path):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::WriteCamerasText(const std::string& path)
+        void Reconstruction::ReadCamerasText(const std::string& path)
+    """
+    cameras = {}
+    with open(path, "r") as fid:
+        while True:
+            line = fid.readline()
+            if not line:
+                break
+            line = line.strip()
+            if len(line) > 0 and line[0] != "#":
+                elems = line.split()
+                camera_id = int(elems[0])
+                model = elems[1]
+                width = int(elems[2])
+                height = int(elems[3])
+                params = np.array(tuple(map(float, elems[4:])))
+                cameras[camera_id] = Camera(id=camera_id, model=model,
+                                            width=width, height=height,
+                                            params=params)
+    return cameras
+
+
+def read_cameras_binary(path_to_model_file=None, fid=None):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::WriteCamerasBinary(const std::string& path)
+        void Reconstruction::ReadCamerasBinary(const std::string& path)
+    """
+    cameras = {}
+    if fid is None:
+        fid = open(path_to_model_file, "rb")
+    num_cameras = read_next_bytes(fid, 8, "Q")[0]
+    for _ in range(num_cameras):
+        camera_properties = read_next_bytes(
+            fid, num_bytes=24, format_char_sequence="iiQQ")
+        camera_id = camera_properties[0]
+        model_id = camera_properties[1]
+        model_name = CAMERA_MODEL_IDS[camera_properties[1]].model_name
+        width = camera_properties[2]
+        height = camera_properties[3]
+        num_params = CAMERA_MODEL_IDS[model_id].num_params
+        params = read_next_bytes(fid, num_bytes=8*num_params,
+                                    format_char_sequence="d"*num_params)
+        cameras[camera_id] = Camera(id=camera_id,
+                                    model=model_name,
+                                    width=width,
+                                    height=height,
+                                    params=np.array(params))
+    assert len(cameras) == num_cameras
+    if path_to_model_file is not None:
+        fid.close()
+    return cameras
+
+
+def write_cameras_text(cameras, path):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::WriteCamerasText(const std::string& path)
+        void Reconstruction::ReadCamerasText(const std::string& path)
+    """
+    HEADER = "# Camera list with one line of data per camera:\n" + \
+             "#   CAMERA_ID, MODEL, WIDTH, HEIGHT, PARAMS[]\n" + \
+             "# Number of cameras: {}\n".format(len(cameras))
+    with open(path, "w") as fid:
+        fid.write(HEADER)
+        for _, cam in cameras.items():
+            to_write = [cam.id, cam.model, cam.width, cam.height, *cam.params]
+            line = " ".join([str(elem) for elem in to_write])
+            fid.write(line + "\n")
+
+
+def write_cameras_binary(cameras, path_to_model_file):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::WriteCamerasBinary(const std::string& path)
+        void Reconstruction::ReadCamerasBinary(const std::string& path)
+    """
+    with open(path_to_model_file, "wb") as fid:
+        write_next_bytes(fid, len(cameras), "Q")
+        for _, cam in cameras.items():
+            model_id = CAMERA_MODEL_NAMES[cam.model].model_id
+            camera_properties = [cam.id,
+                                 model_id,
+                                 cam.width,
+                                 cam.height]
+            write_next_bytes(fid, camera_properties, "iiQQ")
+            for p in cam.params:
+                write_next_bytes(fid, float(p), "d")
+    return cameras
+
+
+def read_images_text(path):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadImagesText(const std::string& path)
+        void Reconstruction::WriteImagesText(const std::string& path)
+    """
+    images = {}
+    with open(path, "r") as fid:
+        while True:
+            line = fid.readline()
+            if not line:
+                break
+            line = line.strip()
+            if len(line) > 0 and line[0] != "#":
+                elems = line.split()
+                image_id = int(elems[0])
+                qvec = np.array(tuple(map(float, elems[1:5])))
+                tvec = np.array(tuple(map(float, elems[5:8])))
+                camera_id = int(elems[8])
+                image_name = elems[9]
+                elems = fid.readline().split()
+                xys = np.column_stack([tuple(map(float, elems[0::3])),
+                                       tuple(map(float, elems[1::3]))])
+                point3D_ids = np.array(tuple(map(int, elems[2::3])))
+                images[image_id] = Image(
+                    id=image_id, qvec=qvec, tvec=tvec,
+                    camera_id=camera_id, name=image_name,
+                    xys=xys, point3D_ids=point3D_ids)
+    return images
+
+
+def read_images_binary(path_to_model_file=None, fid=None):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadImagesBinary(const std::string& path)
+        void Reconstruction::WriteImagesBinary(const std::string& path)
+    """
+    images = {}
+    if fid is None:
+        fid = open(path_to_model_file, "rb")
+    num_reg_images = read_next_bytes(fid, 8, "Q")[0]
+    for _ in range(num_reg_images):
+        binary_image_properties = read_next_bytes(
+            fid, num_bytes=64, format_char_sequence="idddddddi")
+        image_id = binary_image_properties[0]
+        qvec = np.array(binary_image_properties[1:5])
+        tvec = np.array(binary_image_properties[5:8])
+        camera_id = binary_image_properties[8]
+        image_name = ""
+        current_char = read_next_bytes(fid, 1, "c")[0]
+        while current_char != b"\x00":   # look for the ASCII 0 entry
+            image_name += current_char.decode("utf-8")
+            current_char = read_next_bytes(fid, 1, "c")[0]
+        num_points2D = read_next_bytes(fid, num_bytes=8,
+                                        format_char_sequence="Q")[0]
+        x_y_id_s = read_next_bytes(fid, num_bytes=24*num_points2D,
+                                    format_char_sequence="ddq"*num_points2D)
+        xys = np.column_stack([tuple(map(float, x_y_id_s[0::3])),
+                                tuple(map(float, x_y_id_s[1::3]))])
+        point3D_ids = np.array(tuple(map(int, x_y_id_s[2::3])))
+        images[image_id] = Image(
+            id=image_id, qvec=qvec, tvec=tvec,
+            camera_id=camera_id, name=image_name,
+            xys=xys, point3D_ids=point3D_ids)
+    if path_to_model_file is not None:
+        fid.close()
+    return images
+
+
+def write_images_text(images, path):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadImagesText(const std::string& path)
+        void Reconstruction::WriteImagesText(const std::string& path)
+    """
+    if len(images) == 0:
+        mean_observations = 0
+    else:
+        mean_observations = sum((len(img.point3D_ids) for _, img in images.items()))/len(images)
+    HEADER = "# Image list with two lines of data per image:\n" + \
+             "#   IMAGE_ID, QW, QX, QY, QZ, TX, TY, TZ, CAMERA_ID, NAME\n" + \
+             "#   POINTS2D[] as (X, Y, POINT3D_ID)\n" + \
+             "# Number of images: {}, mean observations per image: {}\n".format(len(images), mean_observations)
+
+    with open(path, "w") as fid:
+        fid.write(HEADER)
+        for _, img in images.items():
+            image_header = [img.id, *img.qvec, *img.tvec, img.camera_id, img.name]
+            first_line = " ".join(map(str, image_header))
+            fid.write(first_line + "\n")
+
+            points_strings = []
+            for xy, point3D_id in zip(img.xys, img.point3D_ids):
+                points_strings.append(" ".join(map(str, [*xy, point3D_id])))
+            fid.write(" ".join(points_strings) + "\n")
+
+
+def write_images_binary(images, path_to_model_file):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadImagesBinary(const std::string& path)
+        void Reconstruction::WriteImagesBinary(const std::string& path)
+    """
+    with open(path_to_model_file, "wb") as fid:
+        write_next_bytes(fid, len(images), "Q")
+        for _, img in images.items():
+            write_next_bytes(fid, img.id, "i")
+            write_next_bytes(fid, img.qvec.tolist(), "dddd")
+            write_next_bytes(fid, img.tvec.tolist(), "ddd")
+            write_next_bytes(fid, img.camera_id, "i")
+            for char in img.name:
+                write_next_bytes(fid, char.encode("utf-8"), "c")
+            write_next_bytes(fid, b"\x00", "c")
+            write_next_bytes(fid, len(img.point3D_ids), "Q")
+            for xy, p3d_id in zip(img.xys, img.point3D_ids):
+                write_next_bytes(fid, [*xy, p3d_id], "ddq")
+
+
+def read_points3D_text(path):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadPoints3DText(const std::string& path)
+        void Reconstruction::WritePoints3DText(const std::string& path)
+    """
+    points3D = {}
+    with open(path, "r") as fid:
+        while True:
+            line = fid.readline()
+            if not line:
+                break
+            line = line.strip()
+            if len(line) > 0 and line[0] != "#":
+                elems = line.split()
+                point3D_id = int(elems[0])
+                xyz = np.array(tuple(map(float, elems[1:4])))
+                rgb = np.array(tuple(map(int, elems[4:7])))
+                error = float(elems[7])
+                image_ids = np.array(tuple(map(int, elems[8::2])))
+                point2D_idxs = np.array(tuple(map(int, elems[9::2])))
+                points3D[point3D_id] = Point3D(id=point3D_id, xyz=xyz, rgb=rgb,
+                                               error=error, image_ids=image_ids,
+                                               point2D_idxs=point2D_idxs)
+    return points3D
+
+
+def read_points3D_binary(path_to_model_file=None, fid=None):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadPoints3DBinary(const std::string& path)
+        void Reconstruction::WritePoints3DBinary(const std::string& path)
+    """
+    points3D = {}
+    if fid is None:
+        fid = open(path_to_model_file, "rb")
+    num_points = read_next_bytes(fid, 8, "Q")[0]
+    for _ in range(num_points):
+        binary_point_line_properties = read_next_bytes(
+            fid, num_bytes=43, format_char_sequence="QdddBBBd")
+        point3D_id = binary_point_line_properties[0]
+        xyz = np.array(binary_point_line_properties[1:4])
+        rgb = np.array(binary_point_line_properties[4:7])
+        error = np.array(binary_point_line_properties[7])
+        track_length = read_next_bytes(
+            fid, num_bytes=8, format_char_sequence="Q")[0]
+        track_elems = read_next_bytes(
+            fid, num_bytes=8*track_length,
+            format_char_sequence="ii"*track_length)
+        image_ids = np.array(tuple(map(int, track_elems[0::2])))
+        point2D_idxs = np.array(tuple(map(int, track_elems[1::2])))
+        points3D[point3D_id] = Point3D(
+            id=point3D_id, xyz=xyz, rgb=rgb,
+            error=error, image_ids=image_ids,
+            point2D_idxs=point2D_idxs)
+    if path_to_model_file is not None:
+        fid.close()
+    return points3D
+
+
+def write_points3D_text(points3D, path):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadPoints3DText(const std::string& path)
+        void Reconstruction::WritePoints3DText(const std::string& path)
+    """
+    if len(points3D) == 0:
+        mean_track_length = 0
+    else:
+        mean_track_length = sum((len(pt.image_ids) for _, pt in points3D.items()))/len(points3D)
+    HEADER = "# 3D point list with one line of data per point:\n" + \
+             "#   POINT3D_ID, X, Y, Z, R, G, B, ERROR, TRACK[] as (IMAGE_ID, POINT2D_IDX)\n" + \
+             "# Number of points: {}, mean track length: {}\n".format(len(points3D), mean_track_length)
+
+    with open(path, "w") as fid:
+        fid.write(HEADER)
+        for _, pt in points3D.items():
+            point_header = [pt.id, *pt.xyz, *pt.rgb, pt.error]
+            fid.write(" ".join(map(str, point_header)) + " ")
+            track_strings = []
+            for image_id, point2D in zip(pt.image_ids, pt.point2D_idxs):
+                track_strings.append(" ".join(map(str, [image_id, point2D])))
+            fid.write(" ".join(track_strings) + "\n")
+
+
+def write_points3D_binary(points3D, path_to_model_file):
+    """
+    see: src/base/reconstruction.cc
+        void Reconstruction::ReadPoints3DBinary(const std::string& path)
+        void Reconstruction::WritePoints3DBinary(const std::string& path)
+    """
+    with open(path_to_model_file, "wb") as fid:
+        write_next_bytes(fid, len(points3D), "Q")
+        for _, pt in points3D.items():
+            write_next_bytes(fid, pt.id, "Q")
+            write_next_bytes(fid, pt.xyz.tolist(), "ddd")
+            write_next_bytes(fid, pt.rgb.tolist(), "BBB")
+            write_next_bytes(fid, pt.error, "d")
+            track_length = pt.image_ids.shape[0]
+            write_next_bytes(fid, track_length, "Q")
+            for image_id, point2D_id in zip(pt.image_ids, pt.point2D_idxs):
+                write_next_bytes(fid, [image_id, point2D_id], "ii")
+
+
+def detect_model_format(path, ext):
+    if os.path.isfile(os.path.join(path, "cameras"  + ext)) and \
+       os.path.isfile(os.path.join(path, "images"   + ext)) and \
+       os.path.isfile(os.path.join(path, "points3D" + ext)):
+        print("Detected model format: '" + ext + "'")
+        return True
+
+    return False
+
+
+def read_model(path, ext=""):
+    # try to detect the extension automatically
+    if ext == "":
+        if detect_model_format(path, ".bin"):
+            ext = ".bin"
+        elif detect_model_format(path, ".txt"):
+            ext = ".txt"
+        else:
+            print("Provide model format: '.bin' or '.txt'")
+            return
+
+    if ext == ".txt":
+        cameras = read_cameras_text(os.path.join(path, "cameras" + ext))
+        images = read_images_text(os.path.join(path, "images" + ext))
+        points3D = read_points3D_text(os.path.join(path, "points3D") + ext)
+    else:
+        cameras = read_cameras_binary(os.path.join(path, "cameras" + ext))
+        images = read_images_binary(os.path.join(path, "images" + ext))
+        points3D = read_points3D_binary(os.path.join(path, "points3D") + ext)
+    return cameras, images, points3D
+
+
+def write_model(cameras, images, points3D, path, ext=".bin"):
+    if ext == ".txt":
+        write_cameras_text(cameras, os.path.join(path, "cameras" + ext))
+        write_images_text(images, os.path.join(path, "images" + ext))
+        write_points3D_text(points3D, os.path.join(path, "points3D") + ext)
+    else:
+        write_cameras_binary(cameras, os.path.join(path, "cameras" + ext))
+        write_images_binary(images, os.path.join(path, "images" + ext))
+        write_points3D_binary(points3D, os.path.join(path, "points3D") + ext)
+    return cameras, images, points3D
+
+
+def qvec2rotmat(qvec):
+    return np.array([
+        [1 - 2 * qvec[2]**2 - 2 * qvec[3]**2,
+         2 * qvec[1] * qvec[2] - 2 * qvec[0] * qvec[3],
+         2 * qvec[3] * qvec[1] + 2 * qvec[0] * qvec[2]],
+        [2 * qvec[1] * qvec[2] + 2 * qvec[0] * qvec[3],
+         1 - 2 * qvec[1]**2 - 2 * qvec[3]**2,
+         2 * qvec[2] * qvec[3] - 2 * qvec[0] * qvec[1]],
+        [2 * qvec[3] * qvec[1] - 2 * qvec[0] * qvec[2],
+         2 * qvec[2] * qvec[3] + 2 * qvec[0] * qvec[1],
+         1 - 2 * qvec[1]**2 - 2 * qvec[2]**2]])
+
+
+def rotmat2qvec(R):
+    Rxx, Ryx, Rzx, Rxy, Ryy, Rzy, Rxz, Ryz, Rzz = R.flat
+    K = np.array([
+        [Rxx - Ryy - Rzz, 0, 0, 0],
+        [Ryx + Rxy, Ryy - Rxx - Rzz, 0, 0],
+        [Rzx + Rxz, Rzy + Ryz, Rzz - Rxx - Ryy, 0],
+        [Ryz - Rzy, Rzx - Rxz, Rxy - Ryx, Rxx + Ryy + Rzz]]) / 3.0
+    eigvals, eigvecs = np.linalg.eigh(K)
+    qvec = eigvecs[[3, 0, 1, 2], np.argmax(eigvals)]
+    if qvec[0] < 0:
+        qvec *= -1
+    return qvec

tools2025/hoho2025/vis.py

@@ -0,0 +1,202 @@
+
+import matplotlib.pyplot as plt
+import trimesh
+import numpy as np
+from copy import deepcopy
+from PIL import Image
+
+from . import color_mappings
+
+
+def plot_all_modalities(ds_entry, figsize=(8, 15)):
+    modalities_to_plot = ['images', 'depth', 'gestalt', 'ade']
+    modalities_in_entry = [k for k in ds_entry.keys() if k in modalities_to_plot and len(ds_entry[k]) > 0]
+    number_of_columns = len(modalities_in_entry)
+    number_of_images = len(ds_entry['image_ids'])
+    number_of_rows = number_of_images
+    fig, axes = plt.subplots(number_of_rows, number_of_columns, figsize=figsize)
+    for i in range(len(ds_entry[modalities_in_entry[0]])):
+        for j, modality in enumerate(modalities_in_entry):
+            ax = axes[i, j]
+            if modality == 'image':
+                ax.imshow(ds_entry[modality][i])
+            elif modality == 'depth':
+                depth_image = np.array(ds_entry[modality][i])/1000.0
+                ax.imshow(depth_image, cmap='rainbow')
+            elif modality == 'gestalt':
+                ax.imshow(ds_entry[modality][i])
+            elif modality == 'ade':
+                ax.imshow(ds_entry[modality][i])
+            else:
+                raise ValueError(f"Unknown modality: {modality}")
+            if i == 0:
+                ax.set_title(modality)
+            ax.axis('off')
+            if j == 0:
+                ax.set_ylabel(f"Image {i}")
+    fig.tight_layout()  
+    fig.subplots_adjust(wspace=0.05, hspace=0.01)
+    #plt.show()
+    return fig, axes
+
+
+def line(p1, p2, c=(255,0,0), resolution=10, radius=0.05):
+    '''draws a 3d cylinder along the line (p1, p2)'''
+    # check colors
+    if len(c) == 1:
+        c = [c[0]]*4
+    elif len(c) == 3:
+        c = [*c, 255]
+    elif len(c) != 4:
+        raise ValueError(f'{c} is not a valid color (must have 1,3, or 4 elements).')
+        
+    # compute length and direction of segment
+    p1, p2 = np.asarray(p1), np.asarray(p2)
+    l = np.linalg.norm(p2-p1)
+    
+    direction = (p2 - p1) / l
+    
+    # point z along direction of segment
+    T = np.eye(4)
+    T[:3, 2] = direction
+    T[:3, 3] = (p1+p2)/2
+    
+    #reorthogonalize basis
+    b0, b1 = T[:3, 0], T[:3, 1]
+    if np.abs(np.dot(b0, direction)) < np.abs(np.dot(b1, direction)):
+        T[:3, 1] = -np.cross(b0, direction)
+    else:
+        T[:3, 0] = np.cross(b1, direction)
+    
+    # generate and transform mesh
+    mesh = trimesh.primitives.Cylinder(radius=radius, height=l, transform=T)
+    
+    # apply uniform color
+    mesh.visual.vertex_colors = np.ones_like(mesh.visual.vertex_colors)*c
+         
+    return mesh
+
+def show_wf(row, radius=10, show_vertices=False, vertex_color=(255,0,0, 255)):
+    EDGE_CLASSES = ['eave',
+                    'ridge',
+                    'step_flashing',
+                    'rake',
+                    'flashing',
+                    'post',
+                    'valley',
+                    'hip',
+                    'transition_line']
+    out_meshes = []
+    if show_vertices:
+        out_meshes.extend([trimesh.primitives.Sphere(radius=radius+5, center = center, color=vertex_color) for center in row['wf_vertices']])
+        for m in out_meshes:
+            m.visual.vertex_colors = np.ones_like(m.visual.vertex_colors)*vertex_color
+    if 'edge_semantics' not in row:
+        print ("Warning: edge semantics is not here, skipping")
+        out_meshes.extend([line(a,b, radius=radius, c=(214, 251, 248)) for a,b in np.stack([*row['wf_vertices']])[np.stack(row['wf_edges'])]])
+    elif len(np.stack(row['wf_edges'])) ==  len(row['edge_semantics']):
+        out_meshes.extend([line(a,b, radius=radius, c=color_mappings.gestalt_color_mapping[EDGE_CLASSES[cls_id]]) for (a,b), cls_id in zip(np.stack([*row['wf_vertices']])[np.stack(row['wf_edges'])], row['edge_semantics'])])
+    else:
+        print ("Warning: edge semantics has different length compared to edges, skipping semantics")
+        out_meshes.extend([line(a,b, radius=radius, c=(214, 251, 248)) for a,b in np.stack([*row['wf_vertices']])[np.stack(row['wf_edges'])]])
+    return out_meshes
+    # return [line(a,b, radius=radius, c=color_mappings.edge_colors[cls_id]) for (a,b), cls_id in zip(np.stack([*row['wf_vertices']])[np.stack(row['wf_edges'])], row['edge_semantics'])]
+
+
+def show_grid(edges, meshes=None, row_length=5):
+    '''
+        edges: list of list of meshes
+        meshes: optional corresponding list of meshes
+        row_length: number of meshes per row
+  
+        returns trimesh.Scene()
+    '''
+    
+    T = np.eye(4)
+    out = []
+    edges = [sum(e[1:], e[0]) for e in edges]
+    row_height = 1.1 * max((e.extents for e in edges), key=lambda e: e[1])[1]
+    col_width = 1.1 * max((e.extents for e in edges), key=lambda e: e[0])[0]
+    # print(row_height, col_width)
+    
+    if meshes is None:
+        meshes = [None]*len(edges)
+
+    for i, (gt, mesh) in enumerate(zip(edges, meshes), start=0):
+        mesh = deepcopy(mesh)
+        gt = deepcopy(gt)
+
+        if i%row_length != 0:
+            T[0, 3] += col_width
+
+        else:
+            T[0, 3] = 0
+            T[1, 3] += row_height
+
+        # print(T[0,3]/col_width, T[2,3]/row_height)
+        
+        if mesh is not None:
+            mesh.apply_transform(T)
+            out.append(mesh)
+                            
+        gt.apply_transform(T)
+        out.append(gt)
+        
+                            
+        out.extend([mesh, gt])
+
+            
+    return trimesh.Scene(out)
+
+
+def visualize_order_images(row_order):
+    return create_image_grid(row_order['ade20k'] + row_order['gestalt'] + [visualize_depth(dm) for dm in row_order['depthcm']], num_per_row=len(row_order['ade20k']))
+
+def create_image_grid(images, target_length=312, num_per_row=2):
+    # Calculate the target size for the first image
+    first_img = images[0]
+    aspect_ratio = first_img.width / first_img.height
+    new_width = int((target_length ** 2 * aspect_ratio) ** 0.5)
+    new_height = int((target_length ** 2 / aspect_ratio) ** 0.5)
+    
+    # Resize the first image
+    resized_images = [img.resize((new_width, new_height), Image.Resampling.LANCZOS) for img in images]
+    
+    # Calculate the grid size
+    num_rows = (len(resized_images) + num_per_row - 1) // num_per_row
+    grid_width = new_width * num_per_row
+    grid_height = new_height * num_rows
+    
+    # Create a new image for the grid
+    grid_img = Image.new('RGB', (grid_width, grid_height))
+    
+    # Paste the images into the grid
+    for i, img in enumerate(resized_images):
+        x_offset = (i % num_per_row) * new_width
+        y_offset = (i // num_per_row) * new_height
+        grid_img.paste(img, (x_offset, y_offset))
+    
+    return grid_img
+
+
+def visualize_depth(depth, min_depth=None, max_depth=None, cmap='rainbow'):
+    depth = np.array(depth)
+    
+    if min_depth is None:
+        min_depth = np.min(depth)
+    if max_depth is None:
+        max_depth = np.max(depth)
+    
+    
+    # Normalize the depth to be between 0 and 1
+    depth = (depth - min_depth) / (max_depth - min_depth)
+    depth = np.clip(depth, 0, 1)
+    
+    # Use the matplotlib colormap to convert the depth to an RGB image
+    cmap = plt.get_cmap(cmap)
+    depth_image = (cmap(depth) * 255).astype(np.uint8)
+    
+    # Convert the depth image to a PIL image
+    depth_image = Image.fromarray(depth_image)
+    
+    return depth_image

tools2025/hoho2025/viz3d.py

@@ -0,0 +1,287 @@
+"""
+Copyright [2022] [Paul-Edouard Sarlin and Philipp Lindenberger]
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+
+3D visualization based on plotly.
+Works for a small number of points and cameras, might be slow otherwise.
+
+1) Initialize a figure with `init_figure`
+2) Add 3D points, camera frustums, or both as a pycolmap.Reconstruction
+
+Written by Paul-Edouard Sarlin and Philipp Lindenberger.
+Slightly modified by Dmytro Mishkin
+"""
+from typing import Optional
+import numpy as np
+import pycolmap
+import plotly.graph_objects as go
+from hoho2025.color_mappings import edge_color_mapping, EDGE_CLASSES_BY_ID
+
+def to_homogeneous(points):
+    pad = np.ones((points.shape[:-1]+(1,)), dtype=points.dtype)
+    return np.concatenate([points, pad], axis=-1)
+
+### Plotting functions
+
+def init_figure(height: int = 800) -> go.Figure:
+    """Initialize a 3D figure."""
+    fig = go.FigureWidget()
+    axes = dict(
+        visible=False,
+        showbackground=False,
+        showgrid=False,
+        showline=False,
+        showticklabels=True,
+        autorange=True,
+    )
+    fig.update_layout(
+        template="plotly_dark",
+        height=height,
+        scene_camera=dict(
+            eye=dict(x=0., y=-.1, z=-2),
+            up=dict(x=0, y=-1., z=0),
+            projection=dict(type="orthographic")),
+        scene=dict(
+            xaxis=axes,
+            yaxis=axes,
+            zaxis=axes,
+            aspectmode='data',
+            dragmode='orbit',
+        ),
+        margin=dict(l=0, r=0, b=0, t=0, pad=0),
+        legend=dict(
+            orientation="h",
+            yanchor="top",
+            y=0.99,
+            xanchor="left",
+            x=0.1
+        ),
+    )
+    return fig
+
+
+def plot_lines_3d(
+        fig: go.Figure,
+        pts: np.ndarray,
+        color: str = 'rgba(255, 255, 255, 1)',
+        ps: int = 2,
+        colorscale: Optional[str] = None,
+        name: Optional[str] = None):
+    """Plot a set of 3D points."""
+    x = pts[..., 0]
+    y = pts[..., 1]
+    z = pts[..., 2]
+    if isinstance(color, list):
+        traces = [go.Scatter3d(x=x1, y=y1, z=z1,
+                            mode='lines',
+                            line=dict(color=f"rgb{c}", width=ps)) for x1, y1, z1, c in zip(x,y,z,color)]
+    else:
+        traces = [go.Scatter3d(x=x1, y=y1, z=z1,
+                        mode='lines',
+                        line=dict(color=color, width=ps)) for x1, y1, z1 in zip(x,y,z)]
+    for t in traces:
+        fig.add_trace(t)
+    fig.update_traces(showlegend=False)
+
+
+def plot_points(
+        fig: go.Figure,
+        pts: np.ndarray,
+        color: str = 'rgba(255, 0, 0, 1)',
+        ps: int = 2,
+        colorscale: Optional[str] = None,
+        name: Optional[str] = None):
+    """Plot a set of 3D points."""
+    x, y, z = pts.T
+    tr = go.Scatter3d(
+        x=x, y=y, z=z, mode='markers', name=name, legendgroup=name,
+        marker=dict(
+            size=ps, color=color, line_width=0.0, colorscale=colorscale))
+    fig.add_trace(tr)
+
+def plot_camera(
+        fig: go.Figure,
+        R: np.ndarray,
+        t: np.ndarray,
+        K: np.ndarray,
+        color: str = 'rgb(0, 0, 255)',
+        name: Optional[str] = None,
+        legendgroup: Optional[str] = None,
+        size: float = 1.0):
+    """Plot a camera frustum from pose and intrinsic matrix. R and t are 
+    world_to_camera transformation"""
+    R = np.array(R)
+    t = np.array(t).reshape(3)
+    K = np.array(K)
+    W, H = K[0, 2]*2, K[1, 2]*2
+    corners = np.array([[0, 0], [W, 0], [W, H], [0, H], [0, 0]])
+    if size is not None:
+        image_extent = max(size * W / 1024.0, size * H / 1024.0)
+        world_extent = max(W, H) / (K[0, 0] + K[1, 1]) / 0.5
+        scale = 0.5 * image_extent / world_extent
+    else:
+        scale = 1.0
+    corners = to_homogeneous(corners) @ np.linalg.inv(K).T
+    corners = (corners / 2 * scale) @ R.T + t
+
+    x, y, z = corners.T
+    rect = go.Scatter3d(
+        x=x, y=y, z=z, line=dict(color=color), legendgroup=legendgroup,
+        name=name, marker=dict(size=0.0001), showlegend=False)
+    fig.add_trace(rect)
+
+    x, y, z = np.concatenate(([t], corners)).T
+    i = [0, 0, 0, 0]
+    j = [1, 2, 3, 4]
+    k = [2, 3, 4, 1]
+
+    pyramid = go.Mesh3d(
+        x=x, y=y, z=z, color=color, i=i, j=j, k=k,
+        legendgroup=legendgroup, name=name, showlegend=False)
+    fig.add_trace(pyramid)
+    triangles = np.vstack((i, j, k)).T
+    vertices = np.concatenate(([t], corners))
+    tri_points = np.array([
+        vertices[i] for i in triangles.reshape(-1)
+    ])
+
+    x, y, z = tri_points.T
+    pyramid = go.Scatter3d(
+        x=x, y=y, z=z, mode='lines', legendgroup=legendgroup,
+        name=name, line=dict(color=color, width=1), showlegend=False)
+    fig.add_trace(pyramid)
+
+
+def plot_camera_colmap(
+        fig: go.Figure,
+        image: pycolmap.Image,
+        camera: pycolmap.Camera,
+        name: Optional[str] = None,
+        **kwargs):
+    """Plot a camera frustum from PyCOLMAP objects"""
+    # Use camera intrinsics method if available, otherwise fallback to params
+    intr = camera.calibration_matrix()
+    if intr[0][0] > 5000:
+        print("Bad camera")
+        return
+    world_t_camera = image.cam_from_world.inverse()
+    plot_camera(
+        fig,
+        world_t_camera.rotation.matrix(),  # Use rotation matrix method (World-to-Camera)
+        world_t_camera.translation,  # Use camera center in world coordinates
+        intr,
+        name=name or str(image.name),
+        **kwargs)
+
+
+def plot_cameras(
+        fig: go.Figure,
+        reconstruction: pycolmap.Reconstruction, # Added type hint
+        **kwargs):
+    """Plot a camera as a cone with camera frustum."""
+    # Iterate over reconstruction.images
+    for image_id, image in reconstruction.images.items():
+        # Access camera using reconstruction.cameras
+        plot_camera_colmap(
+            fig, image, reconstruction.cameras[image.camera_id], **kwargs)
+
+
+def plot_reconstruction(
+        fig: go.Figure,
+        rec: pycolmap.Reconstruction, # Added type hint
+        color: str = 'rgb(0, 0, 255)',
+        name: Optional[str] = None,
+        points: bool = True,
+        cameras: bool = True,
+        cs: float = 1.0,
+        single_color_points=False,
+        camera_color='rgba(0, 255, 0, 0.5)',
+        crop_outliers: bool = False):
+    # rec is a pycolmap.Reconstruction object
+    # Filter outliers
+    xyzs = []
+    rgbs = []
+    # Iterate over rec.points3D
+    for k, p3D in rec.points3D.items():
+        #print (p3D)
+        xyzs.append(p3D.xyz)
+        rgbs.append(p3D.color)
+    
+    xyzs = np.array(xyzs)
+    rgbs = np.array(rgbs)
+    
+    # Crop outliers if requested
+    if crop_outliers and len(xyzs) > 0:
+        # Calculate distances from origin
+        distances = np.linalg.norm(xyzs, axis=1)
+        # Find threshold at 98th percentile (removing 2% furthest points)
+        threshold = np.percentile(distances, 98)
+        # Filter points
+        mask = distances <= threshold
+        xyzs = xyzs[mask]
+        rgbs = rgbs[mask]
+        print(f"Cropped outliers: removed {np.sum(~mask)} out of {len(mask)} points ({np.sum(~mask)/len(mask)*100:.2f}%)")
+
+    if points and len(xyzs) > 0:
+        plot_points(fig, xyzs, color=color if single_color_points else rgbs, ps=1, name=name)
+    if cameras:
+        plot_cameras(fig, rec, color=camera_color, legendgroup=name, size=cs)
+
+def plot_wireframe(
+        fig: go.Figure,
+        vertices: np.ndarray,
+        edges: np.ndarray,
+        classifications: np.ndarray = None,
+        color: str = 'rgb(0, 0, 255)',
+        name: Optional[str] = None,
+        **kwargs):
+    """Plot a camera as a cone with camera frustum."""
+    gt_vertices = np.array(vertices)
+    gt_connections = np.array(edges)
+    if gt_vertices is not None:
+        img_color2 = [color for _ in range(len(gt_vertices))]
+        plot_points(fig, gt_vertices, color = img_color2, ps = 10)  
+        if gt_connections is not None:
+            gt_lines = []
+            for c in gt_connections:
+                v1 = gt_vertices[c[0]]
+                v2 = gt_vertices[c[1]]
+                gt_lines.append(np.stack([v1, v2], axis=0))
+            if classifications is not None and len(classifications) == len(gt_lines):
+                line_colors = []
+                for c in classifications:
+                    line_colors.append(edge_color_mapping[EDGE_CLASSES_BY_ID[c]])
+                plot_lines_3d(fig, np.array(gt_lines), line_colors, ps=4)  
+            else:
+                plot_lines_3d(fig, np.array(gt_lines), color, ps=4)  
+
+
+def plot_bpo_cameras_from_entry(fig: go.Figure, entry: dict, idx = None):
+    def cam2world_to_world2cam(R, t):
+        rt = np.eye(4)
+        rt[:3,:3] = R
+        rt[:3,3] = t.reshape(-1)
+        rt = np.linalg.inv(rt)
+        return rt[:3,:3], rt[:3,3]
+    
+    for i in range(len(entry['R'])):
+        if idx is not None and i != idx:
+            continue
+        K = np.array(entry['K'][i])
+        R = np.array(entry['R'][i])
+        t = np.array(entry['t'][i])
+        R, t = cam2world_to_world2cam(R, t)
+        plot_camera(fig, R, t, K)
+    
+    

tools2025/pyproject.toml

@@ -0,0 +1,3 @@
+[build-system]
+requires = ["setuptools>=42", "wheel"]
+build-backend = "setuptools.build_meta"

tools2025/requirements.txt

@@ -0,0 +1,14 @@
+datasets
+huggingface-hub
+ipywidgets
+matplotlib
+numpy
+opencv-python
+Pillow
+plotly
+pycolmap
+scipy
+torch
+trimesh
+webdataset
+manifold3d # for metric computation

tools2025/setup.py

@@ -0,0 +1,37 @@
+from setuptools import setup, find_packages
+import glob
+import os
+
+# Try to read from requirements.txt, but have fallback
+try:
+    here = os.path.abspath(os.path.dirname(__file__))
+    with open(os.path.join(here, 'requirements.txt')) as f:
+        required = f.read().splitlines()
+except FileNotFoundError:
+    # Fallback to hardcoded dependencies
+    required = [
+        'datasets',
+        'huggingface-hub',
+        'ipywidgets',
+        'matplotlib',
+        'numpy',
+        'opencv-python',
+        'Pillow',
+        'plotly',
+        'pycolmap',
+        'scipy',
+        'torch',
+        'trimesh',
+        'webdataset==0.2.111',
+    ]
+
+setup(name='hoho2025',
+	version='0.1.0',
+	description='Tools and utilites for the HoHo Dataset and S23DR Competition',
+	url='https://github.com/s23dr/hoho2025',
+	author='Jack Langerman, Dmytro Mishkin, S23DR Orgainizing Team',
+	author_email='hoho@jackml.com',
+	install_requires=required,
+	packages=find_packages(),
+	python_requires='>=3.10',
+	include_package_data=True)