@@ -211,12 +211,12 @@ def _encode_prompt(
211211
212212 def _run_encoder (
213213 self ,
214- backbone_out ,
215- find_input ,
216- prompt ,
217- prompt_mask ,
214+ backbone_out : Dict ,
215+ find_input : FindStage ,
216+ prompt : torch . Tensor ,
217+ prompt_mask : torch . Tensor ,
218218 encoder_extra_kwargs : Optional [Dict ] = None ,
219- ):
219+ ) -> Tuple [ Dict , Dict , Tuple ] :
220220 feat_tuple = self ._get_img_feats (backbone_out , find_input .img_ids )
221221 backbone_out , img_feats , img_pos_embeds , vis_feat_sizes = feat_tuple
222222
@@ -725,22 +725,22 @@ def __init__(
725725
726726 def forward_video_grounding_multigpu (
727727 self ,
728- backbone_out ,
729- find_inputs ,
728+ backbone_out : Dict ,
729+ find_inputs : List ,
730730 geometric_prompt : Prompt ,
731- frame_idx ,
732- num_frames ,
731+ frame_idx : int ,
732+ num_frames : int ,
733733 # `multigpu_buffer` is a dict to cache detector's outputs in a chunk between different calls
734- multigpu_buffer ,
735- track_in_reverse = False ,
734+ multigpu_buffer : Dict ,
735+ track_in_reverse : bool = False ,
736736 # whether to also return the SAM2 backbone features
737- return_sam2_backbone_feats = False ,
737+ return_sam2_backbone_feats : bool = False ,
738738 # whether to perform NMS and suppress the scores of those detections removed by NMS
739- run_nms = False ,
740- nms_prob_thresh = None ,
741- nms_iou_thresh = None ,
739+ run_nms : bool = False ,
740+ nms_prob_thresh : Optional [ float ] = None ,
741+ nms_iou_thresh : Optional [ float ] = None ,
742742 ** kwargs ,
743- ):
743+ ) -> Tuple [ Dict , Dict ] :
744744 """
745745 Compute the detector's detection outputs in a distributed manner, where all GPUs process
746746 a chunk of frames (equal to the number of GPUs) at once and store them in cache.
0 commit comments