圖3-1為說明HEVC之編碼順序，順序為I frame、P frame

圖3-1為說明HEVC之編碼順序，順序為I frame、P frame、B frame依序編碼，由三種型態的幀組成一組GOP。在多視角編碼時中I frame與P frame為同一幀不同視角，V0為該幀中的I frame，同幀中的V1與V2則是P frame，其餘14幀都是B frame，且編碼順序不會依照幀的順序編碼，但是會在同幀時依照視角順序編碼。 高效率視訊編碼(HEVC)在編碼多視角時預設編碼順序為同幀不同視角，以三視角為例，預設編碼順序為同一幀時的view 0、view 1、view 2依序編碼，利用深度學習生成view 1虛擬圖是利用壓縮後的view 0與view 2，所以我們修改了HM中編碼設定，先讓編碼順序變成view 0、view 2，再進行view 1編碼。在預設中每個B-frame的參考幀有3或4個，但為了加入額外的參考幀，在HM code中修改，限定在B-frame且view 2時加入額外參考幀數目，把額外的參考幀覆蓋上虛擬圖。為了參考一致性，HM原始版本在view 2也多加上額外一參考幀，並以view 1做為跨view的參考幀。

結果 (英文) 1:

Figure 3-1 is an explanatory HEVC coding sequence, the order of I frame, P frame, B frame encoded sequentially by three frames of the type consisting of a set of GOP. When the multi-view encoded I frame and P frame of the same frame different perspective, V0 frame for the I frame, the same frame V1 and V2 is a P frame, the remaining 14 frames are B frame, and the coding sequence does not will, in accordance with the angle of view but will at the same frame encoded sequentially in the order encoded frames.

High-efficiency video coding (HEVC) preset encoding multi-view coding order of frames with a different perspective, an example with three perspective, view 0 is the default coding sequence at the same frame, view 1, view 2 coding sequence, using the generating depth study view 1 is a virtual representation of the view 0 and view 2 by the compression, so we modified the encoded set HM, 0 let coding sequence into view, view 2, view 1 then encoded. In each of the preset reference frame B-frame has 3 or 4, but for adding additional reference frame, modifications HM code, the B-frame is defined when the reference frame number 2 is added and additional View, the additional FIG overlay virtual reference frame. For reference consistency, HM the original version is also more in view 2 plus additional reference frame, and to view 1 as a reference frame of view across.

結果 (英文) 2:

Figure 3-1 shows the coding order of HEVC, which is i frame, P frame, and B frame in sequence, and consists of a set of GOP frames of three patterns. In multi-view encoding, I frame and P frame are different perspectives for the same frame, V0 is the I frame in the frame, V1 and V2 in the same frame are P frames, the remaining 14 frames are B frames, and the encoding order is not encoded in the order of the frames, but will be encoded in the same frame in the order of the viewangle.

High efficiency video coding (HEVC) in coding multi-view angle the default encoding order is different angle of view of the same frame, take three-view angle as an example, the default coding order is the view 0, view 1, view 2 sequential coding at the same frame, using deep learning to generate view 1 virtual graph is to use the compressed view 0 and view view 2, so we modified the coding settings in the HM, first let the coding order become view 0, view 2, and then the view 1 encoding. In the default, there are 3 or 4 reference frames per B-frame, but in order to add additional reference frames, modify in HM code to limit the number of additional reference frames to B-frame 2, and overlay the additional reference frames on the virtual graph. For reference consistency, the original version of HM also adds an additional reference frame to view 2 and uses view 1 as a cross-view reference frame.

結果 (英文) 3:

Figure 3-1 shows the coding sequence of hevc, I frame, P frame and B frame are coded in sequence, and a group of GOPs are composed of three types of frames. In multi view coding, I frame and P frame are the same frame with different views, V0 is the I frame in the frame, V1 and V2 in the same frame are P frame, and the other 14 frames are B frame, and the coding order will not be in accordance with the sequence of frames, but will be in accordance with the sequence of views in the same frame.
High efficiency video coding (hevc) presets the coding order for the same frame and different views when coding multiple views. Take three views as an example, the view 0, view 1 and view 2 are coded in sequence when the coding order is the same frame. The view 1 virtual graph generated by deep learning uses the compressed view 0 and view 2. Therefore, we modified the coding setting in HM to change the coding order into view 0 and view 2, Then view 1 is encoded. In the preset, there are 3 or 4 reference frames for each B-frame, but in order to add additional reference frames, modify it in HM code, limit the number of additional reference frames to be added when B-frame and view 2, and cover the virtual image with additional reference frames. For reference consistency, an additional reference frame is added to view 2 in the original version of HM, and view 1 is used as the cross View reference frame.