effnetb2-b5 test

configs/effb2_meta_default20_extrameta_w_d.yaml

@@ -0,0 +1,30 @@
+MODEL:
+
+  BACKBONE: 'effnetb2'
+  NAME: 'effb2_SVMeta'
+
+  IMG_FCS:  [2816,512]
+
+DATASETS:
+  ROOT_DIR: ['../data/all18_coloradj','../data/all20_coloradj']
+  DICT_LABEL: ['NV', 'MEL'] 
+  INFO_CSV: ['./dat/all18_info1_convert20.csv','./dat/all20_info1.csv']
+  NUM_CLASS: 2
+DATALOADER:
+  BATCH_SIZE: 16
+  SAMPLER: 'weighted_meta'
+  NUM_WORKERS: 8
+TEST:
+  BATCH_SIZE: 32
+
+MISC:
+  OUT_DIR: "../checkpoint/effb2_meta_default20_d"
+  VALID_EPOCH: 10
+  LOG_PERIOD: 10000
+
+INPUT: 
+  SIZE_TRAIN_IN: [260,260]
+  CRP_SCALE: [0.1,1.0]
+  CRP_RATIO: [0.75,1.33]
+SOLVER:
+  EPOCHS: 50

configs/effb2_meta_default20_extrameta_w_d_test.yaml

@@ -0,0 +1,29 @@
+MODEL:
+
+  BACKBONE: 'effnetb2'
+  NAME: 'effb2_SVMeta'
+
+  IMG_FCS:  [2816,512]
+DATASETS:
+  ROOT_DIR: ['../data/all20_usp_coloradj']
+  DICT_LABEL: ['NV', 'MEL'] 
+  INFO_CSV: ['./dat/all20_info_usp.csv']
+  NUM_CLASS: 2
+DATALOADER:
+  BATCH_SIZE: 16
+  SAMPLER: 'weighted_meta'
+  NUM_WORKERS: 8
+TEST:
+  BATCH_SIZE: 32
+
+MISC:
+  OUT_DIR: "../checkpoint/effb2_meta_default20_d"
+  VALID_EPOCH: 10
+  LOG_PERIOD: 10000
+
+INPUT: 
+  SIZE_TRAIN_IN:  [260,260]
+  CRP_SCALE: [0.1,1.0]
+  CRP_RATIO: [0.75,1.33]
+SOLVER:
+  EPOCHS: 50

configs/effb3_meta_default20_extrameta_w_d.yaml

@@ -0,0 +1,30 @@
+MODEL:
+
+  BACKBONE: 'effnetb3'
+  NAME: 'effb3_SVMeta'
+
+  IMG_FCS:  [3072,512]
+
+DATASETS:
+  ROOT_DIR: ['../data/all18_coloradj','../data/all20_coloradj']
+  DICT_LABEL: ['NV', 'MEL'] 
+  INFO_CSV: ['./dat/all18_info1_convert20.csv','./dat/all20_info1.csv']
+  NUM_CLASS: 2
+DATALOADER:
+  BATCH_SIZE: 16
+  SAMPLER: 'weighted_meta'
+  NUM_WORKERS: 8
+TEST:
+  BATCH_SIZE: 32
+
+MISC:
+  OUT_DIR: "../checkpoint/effb3_meta_default20_d"
+  VALID_EPOCH: 10
+  LOG_PERIOD: 10000
+
+INPUT: 
+  SIZE_TRAIN_IN: [300,300]
+  CRP_SCALE: [0.1,1.0]
+  CRP_RATIO: [0.75,1.33]
+SOLVER:
+  EPOCHS: 50

configs/effb3_meta_default20_extrameta_w_d_test.yaml

@@ -0,0 +1,29 @@
+MODEL:
+
+  BACKBONE: 'effnetb3'
+  NAME: 'effb3_SVMeta'
+
+  IMG_FCS:  [3072,512]
+DATASETS:
+  ROOT_DIR: ['../data/all20_usp_coloradj']
+  DICT_LABEL: ['NV', 'MEL'] 
+  INFO_CSV: ['./dat/all20_info_usp.csv']
+  NUM_CLASS: 2
+DATALOADER:
+  BATCH_SIZE: 16
+  SAMPLER: 'weighted_meta'
+  NUM_WORKERS: 8
+TEST:
+  BATCH_SIZE: 32
+
+MISC:
+  OUT_DIR: "../checkpoint/effb3_meta_default20_d"
+  VALID_EPOCH: 10
+  LOG_PERIOD: 10000
+
+INPUT: 
+  SIZE_TRAIN_IN: [300,300]
+  CRP_SCALE: [0.1,1.0]
+  CRP_RATIO: [0.75,1.33]
+SOLVER:
+  EPOCHS: 50

configs/effb5_meta_default20_extrameta_w_d.yaml

@@ -0,0 +1,30 @@
+MODEL:
+
+  BACKBONE: 'effnetb5'
+  NAME: 'effb5_SVMeta'
+
+  IMG_FCS:  [4096,512]
+
+DATASETS:
+  ROOT_DIR: ['../data/all18_coloradj','../data/all20_coloradj']
+  DICT_LABEL: ['NV', 'MEL'] 
+  INFO_CSV: ['./dat/all18_info1_convert20.csv','./dat/all20_info1.csv']
+  NUM_CLASS: 2
+DATALOADER:
+  BATCH_SIZE: 8
+  SAMPLER: 'weighted_meta'
+  NUM_WORKERS: 8
+TEST:
+  BATCH_SIZE: 32
+
+MISC:
+  OUT_DIR: "../checkpoint/effb5_meta_default20_d"
+  VALID_EPOCH: 10
+  LOG_PERIOD: 10000
+
+INPUT: 
+  SIZE_TRAIN_IN: [456,456]
+  CRP_SCALE: [0.1,1.0]
+  CRP_RATIO: [0.75,1.33]
+SOLVER:
+  EPOCHS: 50

configs/effb5_meta_default20_extrameta_w_d_test.yaml

@@ -0,0 +1,29 @@
+MODEL:
+
+  BACKBONE: 'effnetb5'
+  NAME: 'effb5_SVMeta'
+
+  IMG_FCS:  [4096,512]
+DATASETS:
+  ROOT_DIR: ['../data/all20_usp_coloradj']
+  DICT_LABEL: ['NV', 'MEL'] 
+  INFO_CSV: ['./dat/all20_info_usp.csv']
+  NUM_CLASS: 2
+DATALOADER:
+  BATCH_SIZE: 8
+  SAMPLER: 'weighted_meta'
+  NUM_WORKERS: 8
+TEST:
+  BATCH_SIZE: 32
+
+MISC:
+  OUT_DIR: "../checkpoint/effb5_meta_default20_d"
+  VALID_EPOCH: 10
+  LOG_PERIOD: 10000
+
+INPUT: 
+  SIZE_TRAIN_IN: [456,456]
+  CRP_SCALE: [0.1,1.0]
+  CRP_RATIO: [0.75,1.33]
+SOLVER:
+  EPOCHS: 50

convert_18csv_to20_single.py

@@ -0,0 +1,76 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Sat May 30 18:01:58 2020
+use 18 pred result to generate 20 submit table
+@author: cmj
+"""
+import pandas as pd
+import numpy as np
+
+
+label18 = ['MEL', 'NV', 'BCC', 'AKIEC', 'BKL', 'DF', 'VASC']
+
+map_18  = [1,0,1, 1,0,0,0]
+
+
+
+fn_csv_pred = '../checkpoint/18pred_fl.csv'
+fn_csv_out= '../checkpoint/20pred_fl_convert_mc.csv'
+
+meta_fn = '../data/ISIC20/test.csv'
+
+
+pred_18 = pd.read_csv(fn_csv_pred).values
+
+meta_20 = pd.read_csv(meta_fn).values
+
+
+dict_p = dict()
+dict_pn = dict()
+
+
+for pp in pred_18:
+    img_fn = pp[0]
+    pred = pp[1:].astype('float32')
+
+    idx = np.where(meta_20[:,0]==img_fn)[0][0]
+    pid = meta_20[idx,1]
+
+    if pid in dict_p.keys():
+
+        dict_p[pid] += pred
+        dict_pn[pid] +=1
+    else:
+        dict_p[pid] = pred
+        dict_pn[pid] =1
+
+
+
+
+dict_label = [ 'target']
+fn_list = meta_20[:,0]
+targets = list()
+for meta,preds in zip(meta_20,pred_18):
+    fn = meta[0]
+    pid = meta[1]
+
+    pred = preds[1:]
+    p1 = pred[0]+pred[2]+pred[3]
+    p0 = 1-p1
+
+#    p1 = pred[0]
+#    p0 = pred[1]
+#    p1 = p1/(p1+p0)
+
+    #pos = np.argmax(pred)
+    #val = map_18[pos]
+    targets.append(p1)
+
+
+
+
+df = pd.DataFrame(data = targets,index =fn_list, columns = dict_label)
+
+
+
+df.to_csv(fn_csv_out,index_label = 'image_name')