24 lines
910 B
Plaintext
24 lines
910 B
Plaintext
Scripts for evaluating CE-CLM, CLNF, and CLM methods on the 300VW challenge dataset - https://ibug.doc.ic.ac.uk/resources/300-VW/
|
|
|
|
To run the models on the data using models trained on 300W + MultiPIE:
|
|
- Script_CECLM_300VW_general.m
|
|
- Script_CLNF_300VW_general.m
|
|
|
|
To run CE-CLM model trained on 300W + MultiPIE + Menpo:
|
|
- Script_CECLM_300VW_menpo.m
|
|
|
|
To visualize the results of the models against baselines (using inner face markup - 49, and full face markup - 66):
|
|
Display_300VW_results_49.m
|
|
Display_300VW_results_66.m
|
|
|
|
To construct the error table:
|
|
Construct_error_table.m
|
|
|
|
For result computation:
|
|
- Generate_baseline_results.m
|
|
- Generate_ceclm_results.m
|
|
|
|
For a fair comparison all the models and baselines are initialized using the bounding boxes from an MTCNN detector every 30 frames. Other frames use the previous detection to initialize.
|
|
|
|
The best performing model is CE-CLM on 300W + MultiPIE + Menpo.
|