Imported glTF is thumbnail-sized or swallows the camera

• Bounding boxes read 0.62 x 1.87 x 0.55 for model A and 62000 x 187000 x 55000 for model B.
• With one camera setup, A is invisible and B fills the entire scene.
• The model disappears when its extent falls outside camera near/far.

1. Read the bounding box and compare it with the designed real-world size to get the ratio.
2. Print gltf.scene.scale to check whether a parent transform already scales the model.
3. Check camera near/far against the model extent.
4. Frame the model with its bounding sphere before judging whether the size is wrong.

Wrap the scene in a Group and apply a fixed unitScale (0.01 for centimetres, 0.001 for millimetres), or normalize with fitToHeight using Box3 and setScalar, then reposition the camera from the bounding sphere radius.

var box = new THREE.Box3().setFromObject(gltf.scene);
var size = box.getSize(new THREE.Vector3());
console.log('包围盒尺寸', size);
console.log('根节点缩放', gltf.scene.scale.toArray());

• Setting gltf.scene.scale on top of an existing node scale multiplies both and produces a 100x error.
• Guessing 0.01 then 0.001 repeatedly instead of measuring one known dimension wastes time and still misses.