Camera position is not updated after model transform

[zwl995]

My goal was to move the coordinate system to the center of the scene after aligning the arkit, but I found that my camera's center didn't change after aligning, and then I exported the colmap format and found that the model did have conversions. Here is my processing code:

import time
import torch
import pycolmap
import numpy as np
import Metashape
from pathlib import Path

def create_project(output_folder):
    """创建或加载项目（修改为支持加载已有项目）"""
    project_path = Path(output_folder) / "project.psx"
    doc = Metashape.Document()
    if project_path.exists():
        # 加载已有项目（关键：支持增量执行步骤）
        doc.open(str(project_path))
        # 使用第一个chunk（若不存在则新建）
        chunk = doc.chunks[0] if doc.chunks else doc.addChunk()
    else:
        # 创建新项目
        doc.save(str(project_path))
        chunk = doc.addChunk()
    return doc, chunk

def align_scene_center(chunk, doc):
    chunk.crs = Metashape.CoordinateSystem("LOCAL")
    p3ds = []
    for cam in chunk.cameras:
        if cam.center:
            p3ds.append(list(cam.center))
    p3ds = torch.from_numpy(np.array(p3ds).astype(np.float32))
    mean_p3d = torch.mean(p3ds, dim=0)
    print(f"mean_p3d: {mean_p3d}")
    transform2center = Metashape.Matrix.Translation(-mean_p3d.numpy())
    print(f"mean_p3d: {mean_p3d}")
    print(f"transform2center: {transform2center}")
    chunk.transform.matrix = transform2center * chunk.transform.matrix
    # chunk.transform.matrix =  chunk.transform.matrix * transform2center
    chunk.updateTransform()
    return chunk, doc
    
def align_arkit(chunk, doc):
    """对齐ARKIT数据"""
    start = time.time()
    chunk.crs = Metashape.CoordinateSystem("LOCAL")
    for cam in chunk.cameras:
        cam.reference.location = None
        cam.reference.rotation = None
    csv_file = f"/ws/zwl/Data/MetaMaker1160/test_output1/test0904_zhanting/arkit_keyfrm/keyframes_data.csv"
    chunk.importReference(csv_file, columns="nxyz", delimiter=",")
    chunk.updateTransform()
    end = time.time()
    print(f"对齐ARKIT数据耗时: {end - start:.2f} 秒")
    return chunk, doc

if __name__ == "__main__":
    # 创建或加载项目
    output_path = f"/ws/zwl/Data/MetaMaker1160/test_output1/test0904_zhanting/metashape"
    result_folder = f"/ws/zwl/Code/Metashape/Metashape/data"
    doc, chunk = create_project(output_folder=output_path)
    before_align = f"/ws/zwl/Code/Metashape/Metashape/data/before_align/sparse.txt"
    chunk.exportCameras(path=before_align,
                            format=Metashape.CamerasFormatColmap,
                            convert_to_pinhole=True,
                            binary=True,
                            save_images=False,
                            )
    i = 0
    for cam in chunk.cameras:
        if cam.center and i<10:
            print(cam.center)
            i += 1
    chunk, doc = align_arkit(chunk, doc)
    # chunk, doc = align_scene_center(chunk, doc)
    print(f"=============================================================================")
    j = 0
    for cam in chunk.cameras:
        if cam.center and j<10:
            print(cam.center)
            j += 1

    after_align = f"/ws/zwl/Code/Metashape/Metashape/data/after_align_move/align_move.txt"
    chunk.exportCameras(path=after_align,
                        format=Metashape.CamerasFormatColmap,
                        convert_to_pinhole=True,
                        binary=True,
                        save_images=False,
                        )

result is:

[Vineg]

Hi!

You can try this code in python console to move mean camera position to zero coordinates:

chunk = Metashape.app.document.chunk
cameras = chunk.cameras

chunk.crs = Metashape.CoordinateSystem("LOCAL")
for cam in chunk.cameras:
    cam.reference.location = None
    cam.reference.rotation = None

mean_position = Metashape.Vector([0, 0, 0])
cameras_count = 0
for cam in cameras:
    if not cam.enabled or cam.type != Metashape.Camera.Regular:
        continue
    mean_position = mean_position + cam.transform.translation()
    cameras_count += 1

mean_position = mean_position * 1.0 / cameras_count
chunk_translation = - chunk.transform.matrix.mulp(mean_position)
chunk.transform.matrix = Metashape.Matrix.Translation(chunk_translation) * chunk.transform.matrix

[zwl995]

Thank you very much. This will solve my problem.

[zwl995]

你好！

您可以在 Python 控制台中尝试此代码，将平均相机位置移动到零坐标：

chunk = Metashape.app.document.chunk
cameras = chunk.cameras

chunk.crs = Metashape.CoordinateSystem("LOCAL")
for cam in chunk.cameras:
cam.reference.location = None
cam.reference.rotation = None

mean_position = Metashape.Vector([0, 0, 0])
cameras_count = 0
for cam in cameras:
if not cam.enabled or cam.type != Metashape.Camera.Regular:
continue
mean_position = mean_position + cam.transform.translation()
cameras_count += 1

mean_position = mean_position * 1.0 / cameras_count
chunk_translation = - chunk.transform.matrix.mulp(mean_position)
chunk.transform.matrix = Metashape.Matrix.Translation(chunk_translation) * chunk.transform.matrix

Hello, I found that this code not only moves the coordinate system to the center of the scene, but also seems to have rotated it. The image I use is DJI. Before processing the reconstructed model, the SfM model everywhere has the Z-axis facing up. I want to do the following two things:

1. Move coordinate system to scene center
2. Keep the Y axis up
   When I applied the code to complete 1, a strange rotation occurred, as shown below:

[zwl995]

Strangely enough, this did not happen when I aligned Arkit and then moved it to the center of the scene in another task. The model still remained aligned to the axis of the Arkit (Y axis up)