Reproduce value from paper

Hi, thank you for your excellent work!

I'm trying to reproduce the values reported in Table 2 of your paper.
Using the pretrained model and the following command:

```
python -m sample.generate_2stage \
--model_path ./save/diffh2o_full_detailed/model000200000.pt \
--num_samples 16 \
--guidance \
--text_detailed \
--eval_grasp_reference \
--eval_entire_set
```
I evaluated the entire dataset and obtained the following results:
```
iv: 3.6832932692307696
id: 4.474283497497306
id_max: 12.220349985779732
cr: 0.05996021163058345
jp: 0.0012218059557710863
ja: 0.06917452691116986
nc: nan
od: 0.15753278368781132
t_vels: 0.007743637077510357
acc_glob_pos_mean: 0.019779082387685776
acc_loc_pos_mean: 0.006034958176314831
acc_glob_rot_mean: 0.09586863964796066
acc_loc_rot_mean: 0.03712637722492218
handedness_accuracy: 0.9230769230769231
grasp_reference_error: 0.33896538615226746
```
However, several metrics show significant differences compared to Table 2:
| | My Result | Table 2 |
| :-- | :-- | :-- |
| iv | 3.68 | 7.40 |
| ge | 0.34 | 0.12 |
| ha | 0.92 | 0.87 |
| tel | 0.007 | 0.23 |

Did I miss something in the evaluation setup?
Any hints about additional configuration, dataset splits, or evaluation details would be greatly appreciated.

Thanks again for sharing this great work!