test fp16

app.py

@@ -25,8 +25,8 @@ from hls_download import download_clips
 
 plt.style.use('dark_background')
 
-onnx_file = hf_hub_download(repo_id="dylanplummer/ropenet", filename="nextjump.onnx", repo_type="model", token=os.environ['DATASET_SECRET'])
-#onnx_file = hf_hub_download(repo_id="dylanplummer/ropenet", filename="nextjump_fp16.onnx", repo_type="model", token=os.environ['DATASET_SECRET'])
+#onnx_file = hf_hub_download(repo_id="dylanplummer/ropenet", filename="nextjump.onnx", repo_type="model", token=os.environ['DATASET_SECRET'])
+onnx_file = hf_hub_download(repo_id="dylanplummer/ropenet", filename="nextjump_fp16.onnx", repo_type="model", token=os.environ['DATASET_SECRET'])
 # model_xml = hf_hub_download(repo_id="dylanplummer/ropenet", filename="model.xml", repo_type="model", token=os.environ['DATASET_SECRET'])
 # hf_hub_download(repo_id="dylanplummer/ropenet", filename="model.mapping", repo_type="model", token=os.environ['DATASET_SECRET'])
 #model_xml = "model_ir/model.xml"
@@ -134,7 +134,7 @@ def inference(stream_url, start_time, end_time, count_only_api, api_key,
         print('Running inference...')
         if len(batch_list) == batch_size:
             batch_X = torch.cat(batch_list)
-            outputs = ort_sess.run(None, {'video': batch_X.numpy()})
+            outputs = ort_sess.run(None, {'video': np.float16(batch_X.numpy())})
             y1pred = outputs[0]
             y2pred = outputs[1]
             y3pred = outputs[2]
@@ -158,7 +158,7 @@ def inference(stream_url, start_time, end_time, count_only_api, api_key,
             batch_list.append(batch_list[-1])
             idx_list.append(idx_list[-1])
         batch_X = torch.cat(batch_list)
-        outputs = ort_sess.run(None, {'video': batch_X.numpy()})
+        outputs = ort_sess.run(None, {'video': np.float16(batch_X.numpy())})
         y1pred = outputs[0]
         y2pred = outputs[1]
         y3pred = outputs[2]