TLPredictor results missing in domain-shift & disaggregation results/figure/table

[msrepo]

Description

The following figure does not contain results from TLPredictor

Files

hip_outdomain_clinic_metal.json hip_outdomain_clinic.json hip_outdomain_kits.json vertebra_outdomain_rsna.json

benchmark_table.py

Tasks

also, the colour for TL-Embedding and SwinUNETR clash.

benchmark_table not only creates the latex table but also evaluates domain shift results and saves them into json. The functionality is identical for both cases, making it sense to keep it in the table. But on the other hand, it does not hurt to separate them, the functionality common to both may be extracted and put into a utility function somewhere.

• see bash_scripts/evaluate*_embed_decode.sh we need a similar script for domain_shift. currently, generate_embed_decode_arch_evaluation_script.py generates bash scripts for evaluating on benchmark datasets. configs/domain_shift_eval/CTPelvick_abdo

[msrepo]

corrected

The reviewer had correctly identified a mislabelling typo.

It is essential to investigate the reason behind the inconsistent performances of the OneDConcat model between Fig. 5 and Fig. 7. Notably, the performance of OneDConcat appears to be relatively low in Fig. 5, whereas it excels and achieves the best results in Fig. 7.

This updated figure now updates the correct labels.

[msrepo]

commit code for this issue