TheLegendAli / DeepLab-Context2

Other
42 stars 32 forks source link

How to genarate *.bin in FeatureDir #16

Open Hjy20255 opened 7 years ago

Hjy20255 commented 7 years ago

@TheLegendAli in the SetupENV.m is_server = 1;

crf_load_mat = 0; % the densecrf code load MAT files directly (no call SaveMatAsBin.m) % used ONLY by DownSampleFeature.m learn_crf = 0; % NOT USED. Set to 0

is_mat = 0; % the results to be evaluated are saved as mat (1) or png (0) has_postprocess = 1; % has done densecrf post processing (1) or not (0) is_argmax = 0; % the output has been taken argmax already (e.g., coco dataset). % assume the argmax takes C-convention (i.e., start from 0)

debug = 0; % if debug, show some results in the run.py RUN_TRAIN = 0 # Training #1 (on train_aug) RUN_TEST = 0 # Test #1 specification (on val or test) RUN_TRAIN2 = 0 # Training #2 (finetune on trainval_aug) RUN_TEST2 = 0 # Test #2 on official test set RUN_SAVE = 0 # Translate and save the model RUN_DENSECRF = 1 # To Run Densecrf GRID_SEARCH =0 # To Run ONLY if you dont know what parameters to use for Densecrf

i use my own dataset (3 classes) with your deeplabV2 code ,i train successfully. BUT when i run crf, i meet some error which all *.png are black.

Percentage of pixels correctly labelled overall: 51.387% Accuracy for each class (pixel accuracy) background: 100.000% greenhouse: 0.000% plastic_film: 0.000%

Mean Class Accuracy: 33.333% Accuracy for each class (intersection/union measure) background: 51.387% greenhouse: 0.000% plastic_film: 0.000%

Average accuracy: 17.129%