KBA-231226181840
1. Eto Ayika
1.1. Fi Nvidia Driver ati CUDA sori ẹrọ
1.2. Fi Jẹmọ Python Library
python3 -m pip fi sori ẹrọ –igbesoke – foju-fi sori ẹrọ pip
Python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ gdown
Python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ opencv-python
python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ ògùṣọ==1.9.1+cu111 ògùṣọ ògùṣọ==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
Python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ jax
Python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ ftfy
Python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ torchinfo
python3 -m pip fi sori ẹrọ -ignore-fi sori ẹrọ https://github.com/quic/aimet/releases/download/1.25.0/AimetCommon-torch_gpu_1.25.0-cp38-cp38-linux_x86_64.whl
python3 -m pip fi sori ẹrọ -ignore-fi sori ẹrọ https://github.com/quic/aimet/releases/download/1.25.0/AimetTorch-torch_gpu_1.25.0-cp38-cp38-linux_x86_64.whl
Python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ nupy = 1.21.6
python3 -m pip fi sori ẹrọ – foju-fi sori ẹrọ psutil
1.3. Clone aimet-awoṣe-zoo
git clone https://github.com/quic/aimet-model-zoo.git
cd aimet-awoṣe-zoo
git checkout d09d2b0404d10f71a7640a87e9d5e5257b028802
okeere PYTHONPATH=${PYTHONPATH}:${PWD}
1.4. Ṣe igbasilẹ Set14
wget https://uofi.box.com/shared/static/igsnfieh4lz68l926l8xbklwsnnk8we9.zip
unzip igsnfieh4lz68l926l8xbklwsnnk8we9.zip
1.5. Ṣe atunṣe laini 39 aimet-model-zoo/aimet_zoo_torch/quicksrnet/dataloader/utils.py
yipada
fun img_path ni glob.glob(os.path.join(test_images_dir, "*")):
si
fun img_ona ni glob.glob(os.path.join(test_images_dir, “*_HR.*”)):
1.6. Ṣiṣe ayẹwo.
# ṣiṣe labẹ YOURPATH/aimet-model-run
# Fun quicksrnet_small_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-konfigi quicksrnet_small_2x_w8a8 \
-dataset-ona ../Set14/image_SRF_4
# Fun quicksrnet_small_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-konfigi quicksrnet_small_4x_w8a8 \
-dataset-ona ../Set14/image_SRF_4
# Fun quicksrnet_medium_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-konfigi quicksrnet_medium_2x_w8a8 \
-dataset-ona ../Set14/image_SRF_4
# Fun quicksrnet_medium_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
–model-konfigi quicksrnet_medium_4x_w8a8 \
-dataset-ona ../Set14/image_SRF_4
ṣebi iwọ yoo gba PSNRvaluefor theaimetsimulated model. O le yi awoṣe-iṣeto fun iyatọ tiQuickSRNet, aṣayan jẹ underaimet-modelzoo/aimet_zoo_torch/quicksrnet/model/model_cards/.
2 Fi Patch kun
2.1. Ṣii "Firanṣẹ si Awọn Igbesẹ ONNX REVISED.docx"
2.2. Rekọja git dá id
2.3. Abala 1 koodu
Fi odidi 1. koodu labẹ laini to kẹhin (lẹhin laini 366) aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/models.py
2.4. Abala 2 ati 3 koodu
Ṣafikun odidi 2, koodu 3 labẹ laini 93 aimet-model-zoo/aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py
2.5. Awọn paramita bọtini ni iṣẹ load_model
awoṣe = load_model (MODEL_PATH_INT8,
MODEL_NAME,
MODEL_ARGS.gba (MODEL_NAME) .gba (MODEL_CONFIG),
use_quant_sim_model=Otitọ,
encoding_path=ENCODING_PATH,
quantsim_config_path=CONFIG_PATH,
calibration_data=IMAGES_LR,
use_cuda=Òótọ́,
before_quantization=Otitọ,
convert_to_dcr=Otitọ)
MODEL_PATH_INT8 = aimet_zoo_torch/quicksrnet/awoṣe/oṣuwọn/quicksrnet_small_2x_w8a8/pre_opt_weights
MODEL_NAME = QuickSRNetSmall
MODEL_ARGS.gba (MODEL_NAME) .gba (MODEL_CONFIG) = {'scaling_factor': 2}
ENCODING_PATH = aimet_zoo_torch/quicksrnet/awoṣe/oṣuwọn/quicksrnet_small_2x_w8a8/adaround_encodings
CONFIG_PATH = aimet_zoo_torch/quicksrnet/awoṣe/oṣuwọn/quicksrnet_small_2x_w8a8/aimet_config
Jọwọ rọpo awọn oniyipada fun oriṣiriṣi iwọn QuickSRNet
2.6 Awoṣe Iwon Iyipada
- "input_shape" ni aimet-model-zoo/aimet_zoo_torch/quicksrnet/awoṣe/model_cards/*.json
- Iṣẹ inu load_model(...) ni aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/inference.py
- Paramita inu iṣẹ okeere_to_onnx(..., input_height, input_width) lati “Jade lọ si Awọn Igbesẹ ONNX REVISED.docx”
2.7 Tun-Ṣiṣe 1.6 lẹẹkansi fun tajasita awoṣe ONNX
3. Iyipada ni SNPE
3.1. Yipada
${SNPE_ROOT}/bin/x86_64-linux-clang/snpe-onnx-to-dlc \
–input_network model.onnx \
-quantization_overrides ./model.encodings
3.2. (Eyi je ko je) Jade nikan iwon DLC
(iyan) snpe-dlc-quant –input_dlc model.dlc –float_fallback –override_params
3.3. (PATAKI) ONNX I / O wa ni aṣẹ ti NCHW; DLC ti o yipada wa ni aṣẹ NHWC
Awọn iwe aṣẹ / Awọn orisun
 |
Iwe Ohun elo Ohun elo Iṣiṣẹ Qualcomm Aimet [pdf] Awọn ilana quicksrnet_small_2x_w8a8, quicksrnet_small_4x_w8a8, quicksrnet_medium_2x_w8a8, quicksrnet_medium_4x_w8a8, Ohun elo Ohun elo Iṣiṣẹ Aimet, Iwe Ohun elo Iṣẹ ṣiṣe, Iwe Ohun elo Irinṣẹ |
