# Copyright 2026 Toyota Research Institute.  All rights reserved.

import os
import cv2
import numpy as np

from glob import glob

from anydata.utils.read import read_numpy, read_image, read_json
from anydata.utils.write import write_json, write_lowdim, write_labels
from anydata.converters.utils import run, add_key_to_dict, fill_metadata, parse_dst_seq, frame_name, crawl, prepare_lowdim
from anydata.utils.geometry import invert_extrinsics

#######################################################

def get_geometry(transforms, frame):
    idx = int(frame)

    P = transforms["world_mat_" + str(idx)][:3]
    K, R, t = cv2.decomposeProjectionMatrix(P)[:3]
    K = K / K[2, 2]
    K[0, 1] = 0

    pose = np.eye(4, dtype=np.float32)
    pose[:3, :3] = R.transpose()
    pose[:3, 3] = (t[:3] / t[3])[:, 0]

    scale_mtx = transforms.get("scale_mat_" + str(idx))
    if scale_mtx is not None:
        norm_trans = scale_mtx[:3, 3:]
        norm_scale = np.diagonal(scale_mtx[:3, :3])[..., None]
        pose[:3, 3:] -= norm_trans
        pose[:3, 3:] /= norm_scale

    return K, pose

#######################################################

def get_sequences(args):
    seqs = crawl(args.src, 'cameras.npz')
    seqs = [os.path.dirname(seq) for seq in seqs]
    return seqs


def parse_sequence(seq, args):
    return parse_dst_seq(seq, args)

#######################################################

def process_sequence(i, seq, dst, args):

    ### Initialize lists and dicts
    cameras = ['0']
    num_frames = {cam: dict() for cam in cameras}
    resolution = {cam: dict() for cam in cameras}
    labels, lowdim = [], {}
    dense_labels = ['rgb']

    ### Get filenames
    filename_rgbs = sorted(glob(f'{seq}/image/*.png'))

    ### Read transforms for the sequence
    transforms = read_numpy(f'{seq}/cameras.npz')

    ############ LOOP OVER CAMERAS
    for cam in cameras:
        dense = {label: dict() for label in dense_labels}

        ######## RGB FILENAMES
        for i, filename_rgb in enumerate(filename_rgbs):
            frame = frame_name(filename_rgb)

            ######## RGB
            rgb = np.array(read_image(filename_rgb))
            dense['rgb'][frame] = rgb

            ######## LOWDIM RGB             
            prepare_lowdim(lowdim, dst, cam, frame)

            # Get geometry
            intrinsics, extrinsics = get_geometry(transforms, frame)

            ######## INTRINSICS + EXTRINSICS
            filename_lowdim = add_key_to_dict(lowdim, f'{dst}/lowdim/{cam}/{frame}.npz')
            lowdim[filename_lowdim]['extrinsics'] = extrinsics
            lowdim[filename_lowdim]['intrinsics'] = intrinsics

        ######## WRITE LABELS
        write_labels(dst, cam, args.storage, dense, labels, resolution, num_frames)

    ######## WRITE LOWDIM
    write_lowdim(args, dst, labels, num_frames, lowdim)

############ METADATA 
    filename = f'{dst}/metadata.json'
    seq_metadata = fill_metadata(
        args=args,
        info=dict(
            name='DTU',
            tags=['real','static','inward','object'],
            raw_id=seq.replace(f'{args.src}/', ''),
        ),
        labels=labels,
        cameras=cameras,
        resolution=resolution,
        num_frames=num_frames,
        framerate=10,
        rgb=dict(extension='jpg'),
        intrinsics=dict(model='pinhole'),
        extrinsics=dict(transform='cam2world',metric=True),
        depth=None,
        semantic=None,
        action=None,
        language=None,
        specific=None,
    )
    write_json(filename, seq_metadata)

    return dst

#######################################################

if __name__ == '__main__':
    converter = os.path.basename(__file__)
    run(converter, get_sequences, parse_sequence, process_sequence)

#######################################################