[fix] remote anchor available

Author: chyomin06

cfgs/codec/hm.yaml

@@ -29,3 +29,4 @@ encoder_config:
   chroma_format: "400" # "420" for remote inference
   input_bitdepth: 10
   output_bitdepth: 10
+  use_yuv: False

cfgs/codec/jm.yaml

@@ -30,3 +30,4 @@ encoder_config:
   chroma_format: "400" # "420" for remote inference
   input_bitdepth: 10
   output_bitdepth: 10
+  use_yuv: False

cfgs/codec/vtm.yaml

@@ -20,6 +20,7 @@ dump:
   dump_yuv_packing_dec: False
   yuv_packing_dec_dir: ${..output_dir}
   fpn_sizes_json_dump: False
+
 encoder_config:
   qp: 42
   intra_period: 1
@@ -29,3 +30,4 @@ encoder_config:
   chroma_format: "400" # "420" for remote inference
   input_bitdepth: 10
   output_bitdepth: 10
+  use_yuv: False

cfgs/codec/vvenc.yaml

@@ -26,3 +26,4 @@ encoder_config:
   stash_outputs: True
   chroma_format: "420"
   input_bitdepth: 8
+  use_yuv: False

compressai_vision/codecs/encdec_utils/png_yuv.py

@@ -304,7 +304,7 @@ def _determine_output_filename(self, output_file_prefix: str):
         datacatalog = self.datacatalog
 
         # TODO (fracape) hacky, clean this
-        if datacatalog == "MPEGOIV6":
+        if datacatalog in ["MPEGOIV6", "MPEGSAM"]:
             cmd_suffix = []
             filename = f"{output_file_prefix}.png"
         elif datacatalog == "SFUHW":

compressai_vision/model_wrappers/sam.py

@@ -296,7 +296,7 @@ def forward(self, x):
             for k, v in zip(self.split_layer_list, enc_res["data"].values())
         }
 
-        prompts, object_classes = self.prompt_inputs(x["file_name"])
+        prompts, object_classes = self.prompt_inputs(x["file_origin"])
 
         dec_res = self._image_encoder_to_output(
             enc_res["data"],

compressai_vision/pipelines/remote_inference/image_remote_inference.py

@@ -155,7 +155,10 @@ def __call__(
             timing["decode"].append((end - start))
 
             start = time_measure()
-            dec_d = {"file_name": dec_seq["file_names"][0]}
+            dec_d = {
+                "file_name": dec_seq["file_names"][0],
+                "file_origin": d[0]["file_name"],
+            }
             # dec_d = {"file_name": dec_seq[0]["file_names"][0]}
 
             pred = vision_model.forward(org_map_func(dec_d))

compressai_vision/pipelines/remote_inference/video_remote_inference.py

@@ -184,7 +184,10 @@ def __call__(
             # some assertion needed to check if d is matched with dec_seq[e]
 
             start = time_measure()
-            dec_d = {"file_name": dec_seq["file_names"][e]}
+            dec_d = {
+                "file_name": dec_seq["file_names"][e],
+                "file_origin": d[e]["file_name"],
+            }
             # dec_d = {"file_name": dec_seq[0]["file_names"][e]}
             pred = vision_model.forward(org_map_func(dec_d))
             end = time_measure()

compressai_vision/pipelines/split_inference/image_split_inference.py

@@ -198,6 +198,7 @@ def __call__(
                 dec_features["input_size"] = self._get_model_input_size(vision_model, d)
 
             dec_features["file_name"] = d[0]["file_name"]
+            dec_features["file_origin"] = d[0]["file_name"]
             if self.is_mac_calculation:
                 macs, pixels = calc_complexity_nn_part2_plyr(
                     vision_model, dec_features["data"], dec_features

compressai_vision/pipelines/split_inference/video_split_inference.py

@@ -291,6 +291,7 @@ def __call__(
             data = {k: v.to(self.device_nn_part2) for k, v in ftensors.items()}
             dec_features["data"] = data
             dec_features["file_name"] = file_names[e]
+            dec_features["file_origin"] = file_names[e]
             dec_features["qp"] = (
                 "uncmp" if codec.qp_value is None else codec.qp_value
             )  # Assuming one qp will be used