Flow-Developers

Time 2020-12-15 11:04:01

Web Name: Flow-Developers

WebSite: http://flow-developers.hatenablog.com

ID:134870

Keywords:

Flow,Developers,

Description:

Jetson_Convenience_Script/inst_libjpeg-turbo_205.sh at master FREEWING-JP/Jetson_Convenience_Script GitHubwget https://raw.githubusercontent.com/FREEWING-JP/Jetson_Convenience_Script/master/libjpeg-turbo/inst_libjpeg-turbo_205.shchmod 777 inst_libjpeg-turbo_205.sh bash ./inst_libjpeg-turbo_205.sh# specifying library path explicitly# defaultでlinuxの場合は,/usr/lib64/libturbojpeg.soを指してるので指定は不要.# jpeg = TurboJPEG('/usr/lib64/libturbojpeg.so')# using default library installationjpeg = TurboJPEG()# direct rescaling 1/2 while decoding input.jpg to BGR arrayin_file = open('input.jpg', 'rb')# 指定倍率でリサイズ可能. 速い.# supported scaling factors are # [(2, 1), (15, 8), (7, 4), (13, 8), (3, 2), (11, 8), (5, 4), (9, 8), # (1, 1), (7, 8), (3, 4), (5, 8), (1, 2), (3, 8), (1, 4), (1, 8)]bgr_array_half = jpeg.decode(in_file.read(), scaling_factor=(1, 2))in_file.close()cv2.imshow('bgr_array_half', bgr_array_half)cv2.waitKey(0)arm64だと使えなかった高速化ネタpillow-simd, accimage. Desktopで使えば便利そう.DaVinci Resolveで16bitのRGBA(4ch)を扱うためには, FFmpegでRGBA-uint16をProRes4444で圧縮するFFmpegでRGBA-uint16をProRes4444で圧縮することで, DaVinci Resolve上ではfloat32(0.0-1.0[0-65535))の4chで扱えるProRes4444の仕様は,RGB(12bit),A(最大16bit)ですが,DaVinci Resolveは以下のようにProRes4444を書き込むとDaVinci Resolve上ではfloat32(0.0-1.0[0-65535))の4chで扱えました # test frames(16bit gray png) image_test_01 = cv2.imread( image_test_01.png ,0) #16bit_gray_png image_test_02 = cv2.imread( image_test_02.png ,0) #16bit_gray_png image_test_03 = cv2.imread( image_test_03.png ,0) #16bit_gray_png # append array images.append(image_test_01) images.append(image_test_02) images.append(image_test_03) outputfile = rgba_4ch_uint16.mov # ffmpeg prores4444 rgba writer = skvideo.io.FFmpegWriter(outputfile, inputdict={}, outputdict={'-r': str(fps), '-codec:v': 'prores_ks', '-pix_fmt': 'rgba', '-profile:v': '4444', '-bits_per_mb': '8000', '-vendor': 'ap10', '-vf': 'scale=in_color_matrix=bt709:out_color_matrix=bt709'}) # 1ch_16bit to 4ch_16bit for i in range(len(images)): tmp = np.full_like(images[i], 0).astype(np.uint8) rgba = cv2.cvtColor(tmp, cv2.COLOR_GRAY2RGBA) #1ch - 4ch rgba = rgba.astype(np.uint16) #uint8 - uint16 rgba[:,:,0] = np.full_like(images[i], 65535).astype(np.uint16) # R(uint16) rgba[:,:,1] = np.full_like(images[i], 65535).astype(np.uint16) # G(uint16) rgba[:,:,2] = np.full_like(images[i], 65535).astype(np.uint16) # B(uint16) rgba[:,:,3] = images[i].astype(np.uint16) # A(uint16) frame = rgba writer.writeFrame(frame) writer.close()動画を連番png画像で展開(連番で命名)ffmpeg -i input.mp4 -vcodec png image_%03d.pngffmpeg -i input.mp4 -b:v 100M -g 1 -vsync 1 -pix_fmt yuv420p -vcodec libx264 -vf lut3d= 3dlut.cube output.mp4ffmpeg -r ntsc-film -i %03d.png -y -threads 0 -b:v 100M -minrate 100M -maxrate 100M -bufsize 200M -tune film -color_primaries bt709 -color_trc bt709 -colorspace bt709 -g 1 -vsync 1 -pix_fmt yuv422p -vcodec libx264 -r ntsc-film output.mp4ffmpeg -r ntsc-film -start_number 101 -i %03d.png -y -threads 0 -b:v 100M -minrate 100M -maxrate 100M -bufsize 200M -tune film -color_primaries bt709 -color_trc bt709 -colorspace bt709 -g 1 -vsync 1 -pix_fmt yuv422p -vcodec libx264 -r ntsc-film output.mp4ffmpeg -r ntsc-film -i input.mp4 -y -threads 0 -codec:v prores_ks -pix_fmt yuv422p10 -profile:v 3 -bits_per_mb 8000 -vendor ap10 -vf scale=in_color_matrix=bt709:out_color_matrix=bt709 -r ntsc-film output.movffmpeg -r ntsc-film -i input.mp4 -timecode 11:00:06:00 -y -threads 0 -codec:v prores_ks -pix_fmt yuv422p10 -profile:v 3 -bits_per_mb 8000 -vendor ap10 -vf scale=in_color_matrix=bt709:out_color_matrix=bt709 -r ntsc-film output.movffmpeg -r (120*(24/22)) -i input.MP4 -timecode 01:02:03:04 -y -threads 0 -b:v 100M -minrate 100M -maxrate 100M -bufsize 200M -tune film -color_primaries bt709 -color_trc bt709 -colorspace bt709 -g 1 -vsync 1 -pix_fmt yuv420p -vcodec libx264 -r ntsc-film -filter:v setpts=(22/24)*PTS -filter:a atempo=(24/22) -acodec aac -b:a 256k -ar 48000 -vf lut3d= 3dlut.cube output_slow_22p.MP4連番画像(rgba_4chのpng)を23.98fpsのProRes4444で出力ffmpeg -r ntsc-film -i %05d.png -y -threads 0 -codec:v prores_ks -pix_fmt rgba -profile:v 4444 -bits_per_mb 8000 -vendor ap10 -vf scale=in_color_matrix=bt709:out_color_matrix=bt709 -r ntsc-film output.mov連番画像(rgba_4chのpng)を23.98fpsのpng codecで出力ffmpeg -r ntsc-film -i %05d.png -vcodec png -r ntsc-film output_png.movffmpeg -ss 0.6 -i input_prores4444.mov -t 2.0 -y -threads 0 -codec:v prores_ks -pix_fmt rgba -profile:v 4444 -bits_per_mb 8000 -vendor ap10 -vf scale=in_color_matrix=bt709:out_color_matrix=bt709 -r ntsc-film output_prores4444.movM5stickVとKendryte K210が大変興味深かったので,作ってみました.Keras, MobilenetV1のFineTuningで学習しています.Datasetが少ないので,Data Augmetを工夫して,動画の通り満足いく性能が出せました.cd nukkitwget -O nukkit.jar https://ci.nukkitx.com/job/NukkitX/job/Nukkit/job/master/lastSuccessfulBuild/artifact/target/nukkit-1.0-SNAPSHOT.jar Raspberry Pi, Raspbian, python, ROS, IoT, 電子工作, 機械学習, Deep learning, サッカー, スポーツ写真, ライフハック

TAGS:Flow Developers 

<<< Thank you for your visit >>>

Hot Websites