videos_to_colmap.py 24.6 KB
Newer Older
Clement Pinard's avatar
Clement Pinard committed
1
2
3
4
5
6
from colmap_util import read_model as rm, database as db
import anafi_metadata as am
from wrappers import FFMpeg, PDraw
from sklearn.cluster import KMeans
from sklearn.metrics import pairwise_distances_argmin_min
from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter
nicolas's avatar
nicolas committed
7
from edit_exif import set_gps_location
Clement Pinard's avatar
Clement Pinard committed
8
9
10
11
12
from path import Path
import pandas as pd
import numpy as np
from pyproj import Proj
from tqdm import tqdm
13
import tempfile
Clement Pinard's avatar
Clement Pinard committed
14
15

parser = ArgumentParser(description='Take all the drone videos of a folder and put the frame '
Clément Pinard's avatar
Clément Pinard committed
16
                                    'location in a COLMAP file for visualization',
Clement Pinard's avatar
Clement Pinard committed
17
18
19
20
                        formatter_class=ArgumentDefaultsHelpFormatter)

parser.add_argument('--video_folder', metavar='DIR',
                    help='path to videos', type=Path)
Clément Pinard's avatar
Clément Pinard committed
21
22
23
24
25
26
parser.add_argument('--system', default='epsg:2154',
                    help='coordinates system used for GPS, should be the same as the LAS files used')
parser.add_argument('--centroid_path', default=None, help="path to centroid generated in las2ply.py")
parser.add_argument('--colmap_img_root', metavar='DIR', type=Path,
                    help="folder that will be used as \"image_path\" parameter when using COLMAP", required=True)
parser.add_argument('--output_format', metavar='EXT', default="bin", choices=["bin", "txt"],
Clément Pinard's avatar
Clément Pinard committed
27
                    help='format of the COLMAP file that will be outputed, used for visualization only')
Clément Pinard's avatar
Clément Pinard committed
28
29
30
31
parser.add_argument('--vid_ext', nargs='+', default=[".mp4", ".MP4"],
                    help="format of video files that will be scraped from input folder")
parser.add_argument('--pic_ext', nargs='+', default=[".jpg", ".JPG", ".png", ".PNG"],
                    help='format of images that will be scraped from already existing images in colmap image_path folder')
Clement Pinard's avatar
Clement Pinard committed
32
parser.add_argument('--nw', default='',
Clément Pinard's avatar
Clément Pinard committed
33
                    help="native-wrapper.sh file location (see Anafi SDK documentation)")
Clement Pinard's avatar
Clement Pinard committed
34
35
parser.add_argument('--fps', default=1, type=int,
                    help="framerate at which videos will be scanned WITH reconstruction")
Clément Pinard's avatar
Clément Pinard committed
36
37
38
39
40
41
42
parser.add_argument('--total_frames', default=200, type=int, help="number of frames used for thorough photogrammetry")
parser.add_argument('--max_sequence_length', default=1000, help='Number max of frames for a chunk. '
                    'This is for RAM purpose, as loading feature matches of thousands of frames can take up GBs of RAM')
parser.add_argument('--orientation_weight', default=1, type=float,
                    help="Weight applied to orientation during optimal sample. "
                    "Higher means two pictures with same location but different orientation will be considered farer apart")
parser.add_argument('--resolution_weight', default=1, type=float, help="same as orientation, but with image size")
Clément Pinard's avatar
Clément Pinard committed
43
44
parser.add_argument('--save_space', action="store_true",
                    help="if selected, will only extract from ffmpeg frames used for thorough photogrammetry")
Clément Pinard's avatar
Clément Pinard committed
45
parser.add_argument('--thorough_db', type=Path, help="output db file which will be used by COLMAP for photogrammetry")
Clément Pinard's avatar
Clément Pinard committed
46
47
48
parser.add_argument('--generic_model', default='OPENCV',
                    help='COLMAP model for generic videos. Same zoom level assumed throughout the whole video. '
                    'See https://colmap.github.io/cameras.html')
49
50
parser.add_argument('--include_lowfps_thorough', action='store_true',
                    help="if selected, will include videos frames at lowfps for thorough scan, even for generic or indoor videos")
Clément Pinard's avatar
Clément Pinard committed
51
parser.add_argument('-v', '--verbose', action="count", default=0)
Clement Pinard's avatar
Clement Pinard committed
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70


def world_coord_from_frame(frame_qvec, frame_tvec):
    '''
    frame_qvec is written in the NED system (north east down)
    frame_tvec is already is the world system (east norht up)
    '''
    world2NED = np.float32([[0, 1, 0],
                            [1, 0, 0],
                            [0, 0, -1]])
    NED2cam = np.float32([[0, 1, 0],
                          [0, 0, 1],
                          [1, 0, 0]])
    world2cam = NED2cam @ rm.qvec2rotmat(frame_qvec).T @ world2NED
    cam_tvec = - world2cam  @ frame_tvec
    cam_qvec = rm.rotmat2qvec(world2cam)
    return cam_qvec, cam_tvec


71
def set_gps(frames_list, metadata, colmap_img_root):
nicolas's avatar
nicolas committed
72
    for frame in frames_list:
73
        relative = str(frame.relpath(colmap_img_root))
nicolas's avatar
nicolas committed
74
75
76
        row = metadata[metadata["image_path"] == relative]
        if len(row) > 0:
            row = row.iloc[0]
77
            if row["location_valid"] and not row['indoor']:
78
79
80
81
                set_gps_location(frame,
                                 lat=row["location_latitude"],
                                 lng=row["location_longitude"],
                                 altitude=row["location_altitude"])
nicolas's avatar
nicolas committed
82
83
84
85
86
87


def get_georef(metadata):
    relevant_data = metadata[["location_valid", "image_path", "x", "y", "z"]]
    path_list = []
    georef_list = []
Clément Pinard's avatar
Clément Pinard committed
88
    for _, (loc_valid, path, x, y, alt) in relevant_data.iterrows():
nicolas's avatar
nicolas committed
89
        path_list.append(path)
Clément Pinard's avatar
Clément Pinard committed
90
        if loc_valid:
nicolas's avatar
nicolas committed
91
92
93
94
            georef_list.append("{} {} {} {}\n".format(path, x, y, alt))
    return georef_list, path_list


Clement Pinard's avatar
Clement Pinard committed
95
def optimal_sample(metadata, num_frames, orientation_weight, resolution_weight):
96
    # already sampled frames are discarded as we want to sample frames in addition to them
97
    valid_metadata = metadata[~metadata["sampled"]].dropna()
Clément Pinard's avatar
Clément Pinard committed
98
99
    if len(valid_metadata) == 0:
        return metadata
100
101
    XYZ = valid_metadata[["x", "y", "z"]].values
    axis_angle = valid_metadata[["frame_quat_x", "frame_quat_y", "frame_quat_z"]].values
102
    if "indoor" in valid_metadata.keys() and (True in valid_metadata["indoor"].unique()):
103
104
105
        # We have indoor videos, without absolute positions. We assume each video is very far
        # from the other ones. As such we will have an optimal subsampling of each video
        # It won't leverage video proximity from each other but it's better than nothing
Clement Pinard's avatar
Clement Pinard committed
106
        diameter = (XYZ.max(axis=0) - XYZ.min(axis=0))
107
108
109
110
        indoor_videos = valid_metadata.loc[valid_metadata["indoor"]]["video"].unique()
        new_centroids = 2 * diameter * np.linspace(0, 10, len(indoor_videos)).reshape(-1, 1)
        for centroid, v in zip(new_centroids, indoor_videos):
            video_index = (valid_metadata["video"] == v).values
Clement Pinard's avatar
Clement Pinard committed
111
112
113
114
115
116
117
            XYZ[video_index] += centroid

    weighted_point_cloud = np.concatenate([XYZ, orientation_weight * axis_angle], axis=1)

    if resolution_weight == 0:
        weights = None
    else:
118
        frame_size = valid_metadata["video_quality"].values
Clement Pinard's avatar
Clement Pinard committed
119
120
121
        weights = frame_size ** resolution_weight
    km = KMeans(n_clusters=num_frames).fit(weighted_point_cloud, sample_weight=weights)
    closest, _ = pairwise_distances_argmin_min(km.cluster_centers_, weighted_point_cloud)
122
    metadata.at[valid_metadata.index[closest], "sampled"] = True
Clement Pinard's avatar
Clement Pinard committed
123
124
125
    return metadata


126
def register_new_cameras(metadata, device, fields, database, camera_dict):
Clement Pinard's avatar
Clement Pinard committed
127
    camera_ids = []
128
    cameras_dataframe = metadata[metadata["device"] == device][["device"] + fields].drop_duplicates()
Clément Pinard's avatar
Clément Pinard committed
129
    for _, row in cameras_dataframe.iterrows():
130
        camera_model, w, h, params = row.reindex(["camera_model", "width", "height", "camera_params"])
131
132
        model_id = rm.CAMERA_MODEL_NAMES[camera_model].model_id
        num_params = rm.CAMERA_MODEL_NAMES[camera_model].num_params
133
134
135
136
137
138
139
140
141
142
        assert num_params >= len(params), "Got {} params for camera {}".format(len(params), camera_model)
        # Single focal models are SIMPLE_PINHOLE, SIMPLE_RADIAL, SIMPLE_RADIAL_FISHEYE, RADIAL and RADIAL_FISHEYE
        single_focal = ('SIMPLE' in camera_model) or ('RADIAL' in camera_model)
        num_focals = 1 if single_focal else 2
        params = np.array(list(params) + [0] * (num_params - len(params)))

        # prior_focal_length is whether or not COLMAP should rely on it.
        prior_focal_length = all(params[:num_focals] != 0)
        # For unknown focal_length, put a generic placeholder
        params[:num_focals][params[:num_focals] == 0] = w / 2
143
144
145
146
147
148
        # If cx is not set, give the default value w/2
        if params[num_focals] == 0:
            params[num_focals] = w/2
        # same for cy
        if params[num_focals + 1] == 0:
            params[num_focals + 1] = h/2
149
        # We can get less params than actual params if they are unknown. We then pad it with zeros
Clément Pinard's avatar
Clément Pinard committed
150
        db_id = database.add_camera(model_id, int(w), int(h), params, prior_focal_length=prior_focal_length)
Clement Pinard's avatar
Clement Pinard committed
151
152
        camera_ids.append(db_id)
        camera_dict[db_id] = rm.Camera(id=db_id,
153
                                       model=camera_model,
Clement Pinard's avatar
Clement Pinard committed
154
155
156
                                       width=int(w),
                                       height=int(h),
                                       params=params)
157
        metadata.loc[(metadata[["device"] + fields] == row).all(axis=1), "camera_id"] = db_id
Clement Pinard's avatar
Clement Pinard committed
158
159
160
161
    ids_series = pd.Series(camera_ids)
    return cameras_dataframe.set_index(ids_series)


162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
def get_video_metadata(v, output_video_folder, system, generic_model='OPENCV', ** env):
    width, height, framerate, num_frames = env["ffmpeg"].get_size_and_framerate(v)
    video_output_folder = output_video_folder / "{}x{}".format(width, height) / v.stem

    def string_to_tuple(tuple_string):
        assert(tuple_string[0] == '(' and tuple_string[-1] == ')')
        return tuple([float(f) for f in tuple_string[1:-1].split(', ')])

    def generic_metadata():
        metadata = pd.DataFrame({"video": [v] * num_frames})
        metadata["height"] = height
        metadata["width"] = width
        metadata["framerate"] = framerate
        metadata["video_quality"] = height * width / framerate
        metadata['frame'] = metadata.index + 1
        # timestemp is in microseconds
        metadata['time'] = 1e6 * metadata.index / framerate
        metadata['indoor'] = True
        metadata['location_valid'] = False
        metadata["device"] = "generic"
        metadata["camera_model"] = generic_model
        metadata["frame_quat_w"] = np.NaN
        metadata["frame_quat_x"] = np.NaN
        metadata["frame_quat_y"] = np.NaN
        metadata["frame_quat_z"] = np.NaN
        metadata["x"] = np.NaN
        metadata["y"] = np.NaN
        metadata["z"] = np.NaN
        metadata["camera_params"] = [tuple()] * len(metadata)
        return metadata

    # First, try to open the CSV file {video name}_metadata.csv which should contain the metadata
    # If it fails, try to get metadata from MP4 by using PDraw
    # At last resort, simply assume generic parameters

    metadata_file_path = v.parent / "{}_metadata.csv".format(v.stem)
    if metadata_file_path.isfile():
        metadata = pd.read_csv(metadata_file_path)
        # check that the pandas dataframe is well formed
        keys_to_check = ["camera_model", "camera_params", "x", "y", "z",
                         "frame_quat_w", "frame_quat_x", "frame_quat_y", "frame_quat_z",
                         "location_valid", "time"]
        for k in keys_to_check:
            assert k in metadata.keys(), "Metadata file does not contain required field {}".format(k)
        metadata["camera_params"] = metadata["camera_params"].apply(string_to_tuple)
        if "frame" not in metadata.keys():
            metadata["frame"] = range(1, len(metadata) + 1)
        metadata['video'] = v
        if 'indoor' not in metadata.keys():
            metadata['indoor'] = len(metadata[metadata["location_valid"]]) > 0
        if 'video_quality' not in metadata.keys():
            metadata["video_quality"] = height * width / framerate
        device = "other"
    else:
        try:
            proj = Proj(system)
            metadata = am.extract_metadata(v.parent, v, env["pdraw"], proj,
                                           width, height, framerate)
            metadata["camera_model"] = "PINHOLE"
            device = "anafi"
222
        except Exception:
223
224
225
226
227
228
229
230
            # No metadata found, construct a simpler dataframe without location
            metadata = generic_metadata()
            device = "generic"
    metadata["num_frames"] = num_frames
    metadata["device"] = device
    return metadata, device, video_output_folder


231
def process_video_folder(videos_list, individual_pictures, output_video_folder, colmap_img_root, centroid,
Clément Pinard's avatar
Clément Pinard committed
232
                         thorough_db, fps=1, total_frames=500, orientation_weight=1, resolution_weight=1,
233
                         output_colmap_format="bin", save_space=False, include_lowfps_thorough=False,
234
235
                         max_sequence_length=1000, num_neighbours=10,
                         existing_georef=False, existing_metadata=None, **env):
236
    metadata_list = []
Clement Pinard's avatar
Clement Pinard committed
237
238
239
    video_output_folders = {}
    images = {}
    colmap_cameras = {}
240
    tempfile_database = Path(tempfile.NamedTemporaryFile().name)
Clément Pinard's avatar
Clément Pinard committed
241
242
    if thorough_db.isfile():
        thorough_db.copy(thorough_db.stripext() + "_backup.db")
243
244
245
246
247
248
249
250
251
252
    if existing_metadata is not None:
        already_treated_videos = existing_metadata["video"].unique()
        videos_to_treat = [v for v in videos_list if v not in already_treated_videos]
        if len(videos_to_treat) == 0:
            print("All videos already treated. "
                  "Remove the file {} if you want to reprocess everything".format(env["full_metadata"]))
            return None, {}, existing_metadata
        print("Skipping {} already treated videos".format(len(already_treated_videos)))
    else:
        videos_to_treat = videos_list
Clement Pinard's avatar
Clement Pinard committed
253
    path_lists_output = {}
254
    database = db.COLMAPDatabase.connect(thorough_db)
Clement Pinard's avatar
Clement Pinard committed
255
256
    database.create_tables()
    print("extracting metadata for {} videos...".format(len(videos_list)))
257
258
259
    videos_summary = {"anafi": {"indoor": 0, "outdoor": 0},
                      "other": {"indoor": 0, "outdoor": 0},
                      "generic": 0}
260
    for v in tqdm(videos_to_treat):
261
262
263
        metadata, device, output_folder = get_video_metadata(v, output_video_folder, **env)
        video_output_folders[v] = output_folder
        output_folder.makedirs_p()
Clement Pinard's avatar
Clement Pinard committed
264

265
266
267
268
269
270
271
272
273
        if include_lowfps_thorough:
            by_time = metadata.set_index(pd.to_datetime(metadata["time"], unit="us"))
            by_time_lowfps = by_time.resample("{:.3f}S".format(1/fps)).first()
            metadata["sampled"] = by_time["time"].isin(by_time_lowfps["time"]).values
        else:
            metadata["sampled"] = False
        if device == "generic":
            videos_summary["generic"] += 1
        else:
Clément Pinard's avatar
Clément Pinard committed
274
            raw_positions = metadata[["x", "y", "z"]]
275
            if metadata["indoor"].iloc[0]:
276
                videos_summary[device]["indoor"] += 1
277
            else:
278
                videos_summary[device]["outdoor"] += 1
279
            if sum(metadata["location_valid"]) > 0:
280
                if centroid is None:
281
                    '''No centroid (possibly because there was no georeferenced lidar pointcloud in the first place)
282
                    set it as the first valid GPS position of the first outdoor video'''
Clément Pinard's avatar
Clément Pinard committed
283
                    centroid = raw_positions[metadata["location_valid"]].iloc[0].values
284
                zero_centered_positions = raw_positions.values - centroid
Clément Pinard's avatar
bug fix    
Clément Pinard committed
285
286
287
288
                radius = np.max(np.abs(zero_centered_positions))
                if radius > 1000:
                    print("Warning, your positions coordinates are most likely too high, have you configured the right GPS system ?")
                    print("It should be the same as the one used for the Lidar point cloud")
289
                metadata["x"], metadata["y"], metadata["z"] = zero_centered_positions.transpose()
290
291
        metadata_list.append(metadata)
    final_metadata = pd.concat(metadata_list, ignore_index=True)
292
293
    print("{} outdoor anafi videos".format(videos_summary["anafi"]["outdoor"]))
    print("{} indoor anafi videos".format(videos_summary["anafi"]["indoor"]))
294
295
    print("{} indoor other videos".format(videos_summary["other"]["outdoor"]))
    print("{} indoor other videos".format(videos_summary["other"]["indoor"]))
296
    print("{} generic videos".format(videos_summary["generic"]))
Clement Pinard's avatar
Clement Pinard committed
297

298
    if((not existing_georef) and (sum(final_metadata["location_valid"]) == 0) and (videos_summary["anafi"]["indoor"] > 0)):
Clément Pinard's avatar
Clément Pinard committed
299
300
        # We have no GPS data but we have navdata, which will help rescale the colmap model
        # Take the longest video and do as if the GPS was valid
301
302
303
304
305
306
307
308
309
310
311
312
313
314
        indoor_video_diameters = {}
        for md in metadata_list:
            if (metadata["device"].iloc[0] != "anafi") or (not metadata["indoor"].iloc[0]):
                continue
            positions = md[["x", "y", "z"]].values
            video_displacement_diameter = np.linalg.norm(positions.max(axis=0) - positions.min(axis=0))
            if not np.isnan(video_displacement_diameter):
                indoor_video_diameters[video_displacement_diameter] = v

        if len(indoor_video_diameters) > 0:
            longest_video = indoor_video_diameters[max(indoor_video_diameters)]
            print("Only indoor videos used, will use {} for COLMAP rescaling".format(longest_video))
            video_index = final_metadata["video"] == longest_video
            final_metadata.loc[video_index, "location_valid"] = True
Clément Pinard's avatar
Clément Pinard committed
315

Clement Pinard's avatar
Clement Pinard committed
316
317
318
    print("{} frames in total".format(len(final_metadata)))

    final_metadata["camera_id"] = 0
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
    # Set up Anafi cameras, zoom included
    cam_fields = ["camera_model", "width", "height", "camera_params"]
    cam_dfs = []

    if any(final_metadata["device"] == "other"):
        cam_dfs.append(register_new_cameras(final_metadata, "other", cam_fields, database, colmap_cameras))
    if any(final_metadata["device"] == "anafi"):
        # For anafi we don't treat cameras the same if the framerate is different
        # because potentially different rectification algorithms are applied
        anafi_cam_fields = cam_fields + ["framerate"]
        cam_dfs.append(register_new_cameras(final_metadata, "anafi", anafi_cam_fields, database, colmap_cameras))
    if any(final_metadata["device"] == "generic"):
        print("Undefined remaining devices, assigning generic models to them")
        # Fix a single camera per video. This doesn't support different levels of zoom, but
        # COLMAP is not robust to too many different independant camera models
Clément Pinard's avatar
Clément Pinard committed
334
        generic_cam_fields = cam_fields + ["video"]
335
        cam_dfs.append(register_new_cameras(final_metadata, "generic", generic_cam_fields, database, colmap_cameras))
336
    print("Cameras : ")
337
    print(pd.concat(cam_dfs))
338

339
340
    already_sampled = sum(final_metadata["sampled"]) + (existing_metadata["sampled"] if existing_metadata is not None else 0)
    to_extract = total_frames - len(individual_pictures) - already_sampled
Clement Pinard's avatar
Clement Pinard committed
341

Clément Pinard's avatar
Clément Pinard committed
342
    if to_extract <= 0:
343
        pass
Clément Pinard's avatar
Clément Pinard committed
344
    elif to_extract < len(final_metadata):
Clement Pinard's avatar
Clement Pinard committed
345
346
        print("subsampling based on K-Means, to get {}"
              " frames from videos, for a total of {} frames".format(to_extract, total_frames))
347
        final_metadata = optimal_sample(final_metadata, to_extract,
Clement Pinard's avatar
Clement Pinard committed
348
349
350
                                        orientation_weight,
                                        resolution_weight)
        print("Done.")
Clément Pinard's avatar
Clément Pinard committed
351
352
    else:
        final_metadata["sampled"] = True
Clement Pinard's avatar
Clement Pinard committed
353

354
    print("Constructing COLMAP model with {:,} frames".format(sum(final_metadata["sampled"])))
Clement Pinard's avatar
Clement Pinard committed
355

356
357
358
359
    database.commit()
    thorough_db.copy(tempfile_database)
    temp_database = db.COLMAPDatabase.connect(tempfile_database)

Clement Pinard's avatar
Clement Pinard committed
360
    final_metadata["image_path"] = ""
361
362
    final_metadata["db_id"] = -1
    for current_id, row in tqdm(final_metadata.iterrows(), total=len(final_metadata)):
Clement Pinard's avatar
Clement Pinard committed
363
364
365
        video = row["video"]
        frame = row["frame"]
        camera_id = row["camera_id"]
366
        current_image_path = video_output_folders[video].relpath(colmap_img_root) / video.stem + "_{:05d}.jpg".format(frame)
Clement Pinard's avatar
Clement Pinard committed
367

368
369
370
        final_metadata.at[current_id, "image_path"] = current_image_path
        db_image_id = temp_database.add_image(current_image_path, int(camera_id))
        final_metadata.at[current_id, "db_id"] = db_image_id
Clement Pinard's avatar
Clement Pinard committed
371
372
373
374
375
376

        if row["sampled"]:
            frame_qvec = row[["frame_quat_w",
                              "frame_quat_x",
                              "frame_quat_y",
                              "frame_quat_z"]].values
377
378
            if True in pd.isnull(frame_qvec):
                frame_qvec = np.array([1, 0, 0, 0])
nicolas's avatar
nicolas committed
379
380
            x, y, z = row[["x", "y", "z"]]
            frame_tvec = np.array([x, y, z])
381
            if row["location_valid"] and not row['indoor']:
Clement Pinard's avatar
Clement Pinard committed
382
383
384
385
386
                frame_gps = row[["location_longitude", "location_latitude", "location_altitude"]]
            else:
                frame_gps = np.full(3, np.NaN)

            world_qvec, world_tvec = world_coord_from_frame(frame_qvec, frame_tvec)
387
388
389
390
            database.add_image(current_image_path, int(camera_id), prior_t=frame_gps, image_id=db_image_id)
            images[db_image_id] = rm.Image(id=db_image_id, qvec=world_qvec, tvec=world_tvec,
                                           camera_id=camera_id, name=current_image_path,
                                           xys=[], point3D_ids=[])
Clement Pinard's avatar
Clement Pinard committed
391
392
393

    database.commit()
    database.close()
394
395
    temp_database.commit()
    temp_database.close()
Clement Pinard's avatar
Clement Pinard committed
396
397
398
    rm.write_model(colmap_cameras, images, {}, output_video_folder, "." + output_colmap_format)
    print("COLMAP model created")

nicolas's avatar
nicolas committed
399
400
401
402
    thorough_georef, thorough_paths = get_georef(final_metadata[final_metadata["sampled"]])
    path_lists_output["thorough"] = {}
    path_lists_output["thorough"]["frames"] = thorough_paths
    path_lists_output["thorough"]["georef"] = thorough_georef
Clement Pinard's avatar
Clement Pinard committed
403
404
405

    print("Extracting frames from videos")

406
    for v in tqdm(videos_to_treat):
Clement Pinard's avatar
Clement Pinard committed
407
        video_metadata = final_metadata[final_metadata["video"] == v]
nicolas's avatar
nicolas committed
408
        by_time = video_metadata.set_index(pd.to_datetime(video_metadata["time"], unit="us"))
Clement Pinard's avatar
Clement Pinard committed
409
410
        video_folder = video_output_folders[v]
        video_metadata.to_csv(video_folder/"metadata.csv")
nicolas's avatar
nicolas committed
411
412
413
414
415
        path_lists_output[v] = {}
        video_metadata_1fps = by_time.resample("{:.3f}S".format(1/fps)).first()
        georef, frame_paths = get_georef(video_metadata_1fps)
        path_lists_output[v]["frames_lowfps"] = frame_paths
        path_lists_output[v]["georef_lowfps"] = georef
Clément Pinard's avatar
Clément Pinard committed
416
        num_chunks = len(video_metadata) // max_sequence_length + 1
417
418
419
420
        chunks = [list(frames) for frames in np.array_split(video_metadata["image_path"],
                                                            num_chunks)]
        # Add some overlap between chunks, in order to ease the model merging afterwards
        for chunk, next_chunk in zip(chunks, chunks[1:]):
Clément Pinard's avatar
Clément Pinard committed
421
            chunk.extend(next_chunk[:num_neighbours])
422
423
        path_lists_output[v]["frames_full"] = chunks

Clement Pinard's avatar
Clement Pinard committed
424
        if save_space:
Clément Pinard's avatar
Clément Pinard committed
425
426
            frame_ids = set(video_metadata[video_metadata["sampled"]]["frame"].values) | \
                set(video_metadata_1fps["frame"].values)
427
            frame_ids = sorted(list(frame_ids))
Clément Pinard's avatar
Clément Pinard committed
428
            if len(frame_ids) > 0:
nicolas's avatar
nicolas committed
429
                extracted_frames = env["ffmpeg"].extract_specific_frames(v, video_folder, frame_ids)
Clement Pinard's avatar
Clement Pinard committed
430
        else:
nicolas's avatar
nicolas committed
431
            extracted_frames = env["ffmpeg"].extract_images(v, video_folder)
432
433
434
435
        set_gps(extracted_frames, video_metadata, colmap_img_root)
    if existing_metadata is not None:
        final_metadata = pd.concat([existing_metadata, final_metadata], ignore_index=True)
    return path_lists_output, video_output_folders, final_metadata
Clement Pinard's avatar
Clement Pinard committed
436
437
438
439
440
441


if __name__ == '__main__':
    args = parser.parse_args()
    env = vars(args)
    env["videos_list"] = sum((list(args.video_folder.walkfiles('*{}'.format(ext))) for ext in args.vid_ext), [])
Clément Pinard's avatar
Clément Pinard committed
442
    output_video_folder = args.colmap_img_root / "Videos"
Clement Pinard's avatar
Clement Pinard committed
443
444
    output_video_folder.makedirs_p()
    env["output_video_folder"] = output_video_folder
445
    env["individual_pictures"] = sum((list(args.colmap_img_root.walkfiles('*{}'.format(ext))) for ext in args.pic_ext), [])
Clément Pinard's avatar
Clément Pinard committed
446
447
448
    env["pdraw"] = PDraw(args.nw, verbose=args.verbose)
    env["ffmpeg"] = FFMpeg(verbose=args.verbose)
    env["output_colmap_format"] = args.output_format
Clement Pinard's avatar
Clement Pinard committed
449
450
451
452
453
454

    if args.centroid_path is not None:
        centroid = np.loadtxt(args.centroid_path)
    else:
        centroid = np.zeros(3)
    env["centroid"] = centroid
455
    lists, extracted_video_folders, full_metadata = process_video_folder(**env)
Clement Pinard's avatar
Clement Pinard committed
456
457

    if lists is not None:
458
        full_metadata.to_csv(args.colmap_img_root/"full_video_metadata.csv")
Clément Pinard's avatar
Clément Pinard committed
459
460
461
        with open(args.colmap_img_root/"video_frames_for_thorough_scan.txt", "w") as f:
            f.write("\n".join(lists["thorough"]["frames"]) + "\n")
        with open(args.colmap_img_root/"georef.txt", "w") as f:
Clément Pinard's avatar
Clément Pinard committed
462
            f.write("\n".join(lists["thorough"]["georef"]))
Clement Pinard's avatar
Clement Pinard committed
463
        for v in env["videos_list"]:
464
465
            if v not in extracted_video_folders.keys():
                continue
Clément Pinard's avatar
Clément Pinard committed
466
467
468
469
470
471
472
473
474
            video_folder = extracted_video_folders[v]
            with open(video_folder / "lowfps.txt", "w") as f:
                f.write("\n".join(lists[v]["frames_lowfps"]) + "\n")
            with open(video_folder / "georef.txt", "w") as f:
                f.write("\n".join(lists["thorough"]["georef"]) + "\n")
                f.write("\n".join(lists[v]["georef_lowfps"]) + "\n")
            for j, l in enumerate(lists[v]["frames_full"]):
                with open(video_folder / "full_chunk_{}.txt".format(j), "w") as f:
                    f.write("\n".join(l) + "\n")