rename 'ICDAR2017MLT' config & add 'ICDAR2015' config and Model

(cherry picked from commit 4f59eb9)

configs/e2e_PMTD_R_50_FPN_1x_ICDAR2015_test.yaml

@@ -0,0 +1,53 @@
+INPUT:
+  TO_BGR255: False
+  PIXEL_MEAN: [0.485, 0.456, 0.406]
+  PIXEL_STD: [0.229, 0.224, 0.225]
+  MIN_SIZE_TEST: -1
+  MAX_SIZE_TEST: 1920
+TEST:
+  IMS_PER_BATCH: 2
+MODEL:
+  META_ARCHITECTURE: "GeneralizedRCNN"
+  WEIGHT: 'models/PMTD_ICDAR2015.pth'
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+  RESNETS:
+    STEM_FUNC: "StemWithBatchNorm"
+    TRANS_FUNC: "BottleneckWithBatchNorm"
+    BACKBONE_OUT_CHANNELS: 256
+    STRIDE_IN_1X1: False
+  RPN:
+    RPN_HEAD: "SingleConvRPNHead_Softmax"
+    USE_FPN: True
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    ANCHOR_SIZES: (16, 32, 64, 128, 256)
+    ASPECT_RATIOS: (0.17, 0.44, 1.13, 2.90, 7.46)
+    STRADDLE_THRESH: 10
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_PER_BATCH: False
+  ROI_HEADS:
+    USE_FPN: True
+  ROI_BOX_HEAD:
+    NUM_CLASSES: 2
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+  ROI_MASK_HEAD:
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor"
+    PREDICTOR: "MaskRCNNC4Predictor_Upsample"
+    DILATION: 2
+    POOLER_RESOLUTION: 14
+    POOLER_SAMPLING_RATIO: 2
+    RESOLUTION: 28
+    SHARE_BOX_FEATURE_EXTRACTOR: False
+  MASK_ON: True
+DATASETS:
+  TEST: ("icdar_2015_test",)
+DATALOADER:
+  SIZE_DIVISIBILITY: 32

demo/PMTD_demo.py

@@ -51,7 +51,7 @@ def build_parser():
 
 
 def create_pmtd_demo(args):
-    cfg.merge_from_file("configs/e2e_PMTD_R_50_FPN_1x_test.yaml")
+    cfg.merge_from_file("configs/e2e_PMTD_R_50_FPN_1x_ICDAR2017MLT_test.yaml")
     cfg.merge_from_list([
         'MODEL.DEVICE', args.device,
         'MODEL.WEIGHT', args.model_path,

demo/utils/convert_results_to_icdar.py

@@ -1,12 +1,12 @@
 import collections
 import os
-import pyclipper
 import zipfile
 from typing import List, Dict, DefaultDict
 from zipfile import ZipFile
 
 import cv2
 import numpy as np
+import pyclipper
 import torch
 
 
@@ -60,17 +60,11 @@ def inter_of_union(item_a: Dict, item_b: Dict):
     return np.asarray(items)[keep]
 
 
-if __name__ == '__main__':
-    os.chdir('inference/icdar_2017_mlt_test')
-
-    tags = ['1600']
+def output_results(output_dir, scales, image_num, cls_threshold_list, nms_threshold, line_format, pred_file_template):
+    os.chdir(output_dir)
     result_merge_path = 'results_multi_scale.pth'
-    result_path_list = [f'results_{tag}.pth' for tag in tags]
-    results: dict = get_results(result_path_list, result_merge_path, 9000, regenerate=True)
-
-    cls_threshold_list = [0.5]
-    nms_threshold = 0.1
-    items: List[Dict]
+    result_path_list = [f'results_{tag}.pth' for tag in scales]
+    results: dict = get_results(result_path_list, result_merge_path, image_num, regenerate=True)
     with ZipFile('icdar.zip', 'w', compression=zipfile.ZIP_DEFLATED) as result_zip:
         for image_id, item_lists in results.items():
             items = []
@@ -82,6 +76,38 @@ def inter_of_union(item_a: Dict, item_b: Dict):
             lines = []
             for item in items:
                 bbox = item['points'].ravel().astype(np.str).tolist()
-                line = f"{','.join(bbox)}, {item['cls_score']}\n"
+                score = item['cls_score']
+                line = line_format(bbox, score)
                 lines.append(line)
-            result_zip.writestr('res_img_%05d.txt' % image_id, str.join('', lines))
+            result_zip.writestr(pred_file_template % image_id, str.join('', lines))
+
+
+if __name__ == '__main__':
+    # postprocess_config = {
+    #     'output_dir': None,
+    #     'scales': [],
+    #     'image_num': 0,
+    #     'cls_threshold_list': [],
+    #     'nms_threshold': 0,
+    #     'line_format': lambda bbox, cls_score: None,
+    #     'pred_file_template': None
+    # }
+    postprocess_config_17 = {
+        'output_dir': 'inference/icdar_2017_mlt_test',
+        'scales': [1600],
+        'image_num': 9000,
+        'cls_threshold_list': [0.5],
+        'nms_threshold': 0.1,
+        'line_format': lambda bbox, cls_score: f"{','.join(bbox)}, {cls_score}\n",
+        'pred_file_template': 'res_img_%05d.txt'
+    }
+    postprocess_config_15 = {
+        'output_dir': 'inference/icdar_2015_test',
+        'scales': [1920],
+        'image_num': 500,
+        'cls_threshold_list': [0.4],
+        'nms_threshold': 0.3,
+        'line_format': lambda bbox, cls_score: f"{','.join(bbox)}\n",
+        'pred_file_template': 'res_img_%d.txt'
+    }
+    output_results(**postprocess_config_17)

maskrcnn_benchmark/config/paths_catalog.py

@@ -104,6 +104,10 @@ class DatasetCatalog(object):
             "img_dir": "cityscapes/images",
             "ann_file": "cityscapes/annotations/instancesonly_filtered_gtFine_test.json"
         },
+        "icdar_2015_test": {
+            "img_dir": "icdar2015/ch4_test_images",
+            "ann_file": "icdar2015/annotations/test_coco.json"
+        },
         "icdar_2017_mlt_val": {
             "img_dir": "icdar2017mlt/ch8_validation_images",
             "ann_file": "icdar2017mlt/annotations/validate_coco_with_ignore.json"