fix rtdetr yaml and infer (#8268)

deploy/auto_compression/configs/rtdetr_hgnetv2_x_qat_dis.yaml

@@ -3,7 +3,7 @@ Global:
   reader_config: configs/rtdetr_reader.yml
   include_nms: True
   Evaluation: True
-  model_dir: ./rtdetr_r50vd_6x_coco/ 
+  model_dir: ./rtdetr_hgnetv2_x_6x_coco/
   model_filename: model.pdmodel
   params_filename: model.pdiparams
 

deploy/auto_compression/configs/rtdetr_r101vd_qat_dis.yaml

@@ -3,7 +3,7 @@ Global:
   reader_config: configs/rtdetr_reader.yml
   include_nms: True
   Evaluation: True
-  model_dir: ./rtdetr_hgnetv2_x_6x_coco/
+  model_dir: ./rtdetr_r101vd_6x_coco/ 
   model_filename: model.pdmodel
   params_filename: model.pdiparams
 

deploy/auto_compression/configs/rtdetr_reader.yml

@@ -12,6 +12,18 @@ TrainDataset:
     anno_path: annotations/instances_val2017.json
     dataset_dir: dataset/coco/
 
+EvalDataset:
+  !COCODataSet
+    image_dir: val2017
+    anno_path: annotations/instances_val2017.json
+    dataset_dir: dataset/coco/
+
+TestDataset:
+  !COCODataSet
+    image_dir: val2017
+    anno_path: annotations/instances_val2017.json
+    dataset_dir: dataset/coco/
+
 worker_num: 0
 
 # preprocess reader in test

deploy/auto_compression/paddle_inference_eval.py

@@ -284,48 +284,6 @@ def load_predictor(
     return predictor, rerun_flag
 
 
-def get_current_memory_mb():
-    """
-    It is used to Obtain the memory usage of the CPU and GPU during the running of the program.
-    And this function Current program is time-consuming.
-    """
-    try:
-        pkg.require('pynvml')
-    except:
-        from pip._internal import main
-        main(['install', 'pynvml'])
-    try:
-        pkg.require('psutil')
-    except:
-        from pip._internal import main
-        main(['install', 'psutil'])
-    try:
-        pkg.require('GPUtil')
-    except:
-        from pip._internal import main
-        main(['install', 'GPUtil'])
-    import pynvml
-    import psutil
-    import GPUtil
-
-    gpu_id = int(os.environ.get("CUDA_VISIBLE_DEVICES", 0))
-
-    pid = os.getpid()
-    p = psutil.Process(pid)
-    info = p.memory_full_info()
-    cpu_mem = info.uss / 1024.0 / 1024.0
-    gpu_mem = 0
-    gpu_percent = 0
-    gpus = GPUtil.getGPUs()
-    if gpu_id is not None and len(gpus) > 0:
-        gpu_percent = gpus[gpu_id].load
-        pynvml.nvmlInit()
-        handle = pynvml.nvmlDeviceGetHandleByIndex(0)
-        meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
-        gpu_mem = meminfo.used / 1024.0 / 1024.0
-    return round(cpu_mem, 4), round(gpu_mem, 4)
-
-
 def predict_image(predictor,
                   image_file,
                   image_shape=[640, 640],
@@ -367,13 +325,7 @@ def predict_image(predictor,
         time_min = min(time_min, timed)
         time_max = max(time_max, timed)
         predict_time += timed
-        cpu_mem, gpu_mem = get_current_memory_mb()
-        cpu_mems += cpu_mem
-        gpu_mems += gpu_mem
-
     time_avg = predict_time / repeats
-    print("[Benchmark]Avg cpu_mem:{} MB, avg gpu_mem: {} MB".format(
-        cpu_mems / repeats, gpu_mems / repeats))
     print("[Benchmark]Inference time(ms): min={}, max={}, avg={}".format(
         round(time_min * 1000, 2),
         round(time_max * 1000, 1), round(time_avg * 1000, 1)))
@@ -418,9 +370,6 @@ def eval(predictor, val_loader, metric, rerun_flag=False):
         time_min = min(time_min, timed)
         time_max = max(time_max, timed)
         predict_time += timed
-        cpu_mem, gpu_mem = get_current_memory_mb()
-        cpu_mems += cpu_mem
-        gpu_mems += gpu_mem
         if not FLAGS.include_nms:
             postprocess = PPYOLOEPostProcess(
                 score_threshold=0.3, nms_threshold=0.6)
@@ -436,8 +385,6 @@ def eval(predictor, val_loader, metric, rerun_flag=False):
     map_res = metric.get_results()
     metric.reset()
     time_avg = predict_time / sample_nums
-    print("[Benchmark]Avg cpu_mem:{} MB, avg gpu_mem: {} MB".format(
-        cpu_mems / sample_nums, gpu_mems / sample_nums))
     print("[Benchmark]Inference time(ms): min={}, max={}, avg={}".format(
         round(time_min * 1000, 2),
         round(time_max * 1000, 1), round(time_avg * 1000, 1)))