[dev] fix summarize_performance in the remote pipeline

Author: chyomin06

compressai_vision/run/eval_remote_inference.py

@@ -170,7 +170,7 @@ def main(conf: DictConfig):
     evaluator_name = _get_evaluator_name(**modules)
     evaluator_filepath = _get_evaluator_filepath(**modules)
     seq_info_path = _get_seqinfo_path(**modules)
-    performance, eval_criteria = _summerize_performance(
+    performance, eval_criteria = _summarize_performance(
         evaluator_name, performance, conf.evaluator.eval_criteria
     )
 
@@ -230,7 +230,8 @@ def _calc_bpp(coded_res_df):
     return avg_bpp
 
 
-def _summerize_performance(evaluator_name, performance, eval_criteria):
+def _summarize_performance(evaluator_name, performance, eval_criteria):
+    # Factorization needed TODO (Hyomin)
     if evaluator_name == "OpenImagesChallengeEval":
         def_criteria = "[email protected]"
         if not eval_criteria:
@@ -243,8 +244,7 @@ def _summerize_performance(evaluator_name, performance, eval_criteria):
             eval_criteria = def_criteria
             value = [v for k, v in performance.items() if k.endswith(eval_criteria)]
         return value, eval_criteria
-
-    if evaluator_name == "COCOEVal":
+    elif evaluator_name == "COCOEVal":
         def_criteria = "AP"
         if not eval_criteria:
             eval_criteria = def_criteria
@@ -256,8 +256,7 @@ def _summerize_performance(evaluator_name, performance, eval_criteria):
             eval_criteria = def_criteria
             value = [v for k, v in performance["bbox"].items() if k == eval_criteria]
         return value, eval_criteria
-
-    if evaluator_name == "MOT_TVD_Eval" or evaluator_name == "MOT_HiEve_Eval":
+    elif evaluator_name == "MOT_TVD_Eval" or evaluator_name == "MOT_HiEve_Eval":
         def_criteria = "mota"
         if not eval_criteria:
             eval_criteria = def_criteria
@@ -269,6 +268,21 @@ def _summerize_performance(evaluator_name, performance, eval_criteria):
             eval_criteria = def_criteria
             value = [v for k, v in performance.items() if k == eval_criteria]
         return value, eval_criteria
+    elif evaluator_name == "YOLOXCOCOEval":
+        def_criteria = "AP"
+        if not eval_criteria:
+            eval_criteria = def_criteria
+        value = [v for k, v in performance.items() if k == eval_criteria]
+        if not value:
+            print(
+                f"\n{eval_criteria} is not supported for {evaluator_name}, using default evaluation criteria {def_criteria}"
+            )
+            eval_criteria = def_criteria
+            value = [v for k, v in performance.items() if k == eval_criteria]
+        return value, eval_criteria
+    else:
+        raise NotImplementedError
+
     return performance, eval_criteria
 
 

compressai_vision/run/eval_split_inference.py

@@ -247,7 +247,7 @@ def main(conf: DictConfig):
 
             elap_times = updates
 
-    performance, eval_criteria = _summerize_performance(
+    performance, eval_criteria = _summarize_performance(
         evaluator_name, performance, conf.evaluator.eval_criteria
     )
 
@@ -326,7 +326,8 @@ def _calc_bpp(coded_res_df):
     return avg_bpp
 
 
-def _summerize_performance(evaluator_name, performance, eval_criteria):
+def _summarize_performance(evaluator_name, performance, eval_criteria):
+    # Factorization needed TODO (Hyomin)
     if evaluator_name == "OpenImagesChallengeEval":
         def_criteria = "[email protected]"
         if not eval_criteria: