KBA-231226181840
1. Seta Tikoloho
1.1. Kenya Nvidia Driver le CUDA
1.2. Kenya Laebrari e amanang le Python
python3 -m pip install -upgrade -ignore-installed pip
python3 -m pip install -ignore-installed gdown
python3 -m pip install - hlokomoloha-o kentsoeng opencv-python
python3 -m pip install -ignore-installed torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
python3 -m pip install -ho hlokomoloha jax e kentsoeng
python3 -m pip install - hlokomoloha-e kentsoeng ftfy
python3 -m pip install - hlokomoloha-torchinfo e kentsoeng
python3 -m pip kenya -ignore-e kentsoe 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 install -ignore-e kentsoe 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 install -ignore-e kentse numpy ==1.21.6
python3 -m pip install -ignore-installed psutil
1.3. Clone aimet-mohlala-zoo
git clone https://github.com/quic/aimet-model-zoo.git
cd aimet-mohlala-zoo
git checkout d09d2b0404d10f71a7640a87e9d5e5257b028802
romela kantle PYTHONPATH=${PYTHONPATH}:${PWD}
1.4. Khoasolla Set14
wget https://uofi.box.com/shared/static/igsnfieh4lz68l926l8xbklwsnnk8we9.zip
unzip igsnfieh4lz68l926l8xbklwsnnk8we9.zip
1.5. Fetola mola 39 aimet-model-zoo/aimet_zoo_torch/quicksrnet/dataloader/utils.py
phetoho
bakeng sa img_path ho glob.glob(os.path.join(test_images_dir, “*”)):
ho
bakeng sa img_path ho glob.glob(os.path.join(test_images_dir, “*_HR.*”)):
1.6. Etsa tlhahlobo.
# matha tlasa YOURPATH/aimet-model-run
# Bakeng sa quicksrnet_small_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_small_2x_w8a8 \
-dataset-path ../Set14/image_SRF_4
# Bakeng sa quicksrnet_small_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_small_4x_w8a8 \
-dataset-path ../Set14/image_SRF_4
# Bakeng sa quicksrnet_medium_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_medium_2x_w8a8 \
-dataset-path ../Set14/image_SRF_4
# Bakeng sa quicksrnet_medium_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_medium_4x_w8a8 \
-dataset-path ../Set14/image_SRF_4
ha re re u tla fumana PSNRvaluefor theaimetsimulated model. O ka fetola sebopeho-sebopeho bakeng sa boholo bo fapaneng baQuickSRNet, khetho ke underaimet-modelzoo/aimet_zoo_torch/quicksrnet/model/model_cards/.
2 Eketsa Patch
2.1. Bula "Export to ONNX Steps REVISED.docx"
2.2. Tlola id ea boitlamo ba git
2.3. Karolo ea 1 Khoutu
Kenya khoutu e felletseng ea 1 tlas'a mola oa ho qetela (kamora mola oa 366) aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/models.py
2.4. Karolo ea 2 le ea 3 Khoutu
Kenya khoutu e felletseng ea 2, 3 tlasa mola oa 93 aimet-model-zoo/aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py
2.5. Lintlha tsa bohlokoa ho Function load_model
mohlala = load_model(MODEL_PATH_INT8,
MODEL_NAME,
MODEL_ARGS.fumana(MODEL_NAME).fumana(MODEL_CONFIG),
use_quant_sim_model=Ke 'nete,
encoding_path=ENCODING_PATH,
quantsim_config_path=CONFIG_PATH,
calibration_data=IMAGES_LR,
use_cuda=Nnete,
before_quantization=Nnete,
convert_to_dcr=Nnete)
MODEL_PATH_INT8 = aimet_zoo_torch/quicksrnet/model/weights/quicksrnet_small_2x_w8a8/pre_opt_weights
MODEL_NAME = QuickSRNetSmall
MODEL_ARGS.get(MODEL_NAME).fumana(MODEL_CONFIG) = {'scaling_factor': 2}
ENCODING_PATH = aimet_zoo_torch/quicksrnet/model/weights/quicksrnet_small_2x_w8a8/adaround_encodings
CONFIG_PATH = aimet_zoo_torch/quicksrnet/model/weights/quicksrnet_small_2x_w8a8/aimet_config
Ka kopo, nkela sebaka sa mefuta-futa bakeng sa boholo bo fapaneng ba QuickSRNet
2.6 Phetoho ea Boholo ba Mohlala
- “input_shape” ho aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/model_cards/*.json
- Ka hare ho tshebetso load_model(…) ho aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/inference.py
- Paramethara ka hare ho tshebetso export_to_onnx(…, input_height, input_width) ho tloha "Export to ONNX Steps REVISED.docx"
2.7 Etsa hape 1.6 hape bakeng sa ho romela mohlala oa ONNX kantle ho naha
3. Fetolela ho SNPE
3.1. Fetolela
${SNPE_ROOT}/bin/x86_64-linux-clang/snpe-onnx-to-dlc \
-input_network model.onnx \
–quantization_overrides ./model.encodings
3.2. (Ka boikhethelo) Ntša feela DLC e lekantsoeng
(ka boikhethelo) snpe-dlc-quant -input_dlc model.dlc -float_fallback -override_params
3.3. (BOHLOKOA) ONNX I/O e latela tatellano ea NCHW; DLC e fetotsoeng e maemong a NHWC
Litokomane / Lisebelisoa
 |
Qualcomm Aimet Efficiency Toolkit Documentation [pdf] Litaelo quicksrnet_small_2x_w8a8, quicksrnet_small_4x_w8a8, quicksrnet_medium_2x_w8a8, quicksrnet_medium_4x_w8a8, Aimet Efficiency Toolkit Documentation, Efficiency Toolkit Documentation, Toolkit Document |
