KBA-231226181840
1. Nhazi gburugburu
1.1. Wụnye Nvidia Driver na CUDA
1.2. Wụnye ọba akwụkwọ Python emetụtara
python3 -m pip wụnye -upgrade -eleghara pip arụnyere
python3 -m pip wụnye -eleghara gdown arụnyere
python3 -m pip wụnye-eleghara-arụnyere opencv-python
python3 -m pip wụnye-eleghara ọwa arụnyere = 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 wụnye-eleghara jax arụnyere
python3 -m pip wụnye-eleghara arụnyere ftfy
python3 -m pip wụnye-eleghara-ịwụnye torchinfo
python3 -m pip wụnye -ignore-installed 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 wụnye -ignore-installed 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 wụnye-eleghara nọmba arụnyere arụnyere = 1.21.6
python3 -m pip wụnye-eleghara-arụnyere psutil
1.3. Clone aimet-model-zoo
git clone https://github.com/quic/aimet-model-zoo.git
cd aimet-model-zoo
git checkout d09d2b0404d10f71a7640a87e9d5e5257b028802
mbupu PYTHONPATH=${PYTHONPATH}:${PWD}
1.4. Budata Set14
wget https://uofi.box.com/shared/static/igsnfieh4lz68l926l8xbklwsnnk8we9.zip
unzip igsnfieh4lz68l926l8xbklwsnnk8we9.zip
1.5. Gbanwee ahịrị 39 aimet-model-zoo/aimet_zoo_torch/quicksrnet/dataloader/utils.py
gbanwee
maka img_ụzọ na glob.glob(os.path.join(test_images_dir, "*")):
ka
maka img_ụzọ na glob.glob(os.path.join(test_images_dir, "*_HR.*")):
1.6. Gbaa nyocha.
# na-agba ọsọ n'okpuru YOURPATH/aimet-model-run
# Maka quicksrnet_small_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_small_2x_w8a8 \
-ụzọ dataset ../Set14/image_SRF_4
# Maka quicksrnet_small_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_small_4x_w8a8 \
-ụzọ dataset ../Set14/image_SRF_4
# Maka quicksrnet_medium_2x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_medium_2x_w8a8 \
-ụzọ dataset ../Set14/image_SRF_4
# Maka quicksrnet_medium_4x_w8a8
python3 aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py \
-model-config quicksrnet_medium_4x_w8a8 \
-ụzọ dataset ../Set14/image_SRF_4
were were na ị ga-enweta PSNRvalue maka ihe ngosi aimetsimulated. Ị nwere ike ịgbanwe ụdị-nhazi maka dị iche iche nkeQuickSRNet, nhọrọ bụ underaimet-modelzoo/aimet_zoo_torch/quicksrnet/model/model_cards/.
2 Tinye patch
2.1. Mepee "Bupu gaa na ONNX nzọụkwụ REVISED.docx"
2.2. Mafere git eme id
2.3. Koodu ngalaba 1
Tinye dum 1. koodu n'okpuru ikpeazụ ahịrị (mgbe akara 366 gasịrị) aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/models.py
2.4. Usoro 2 na 3
Tinye 2, 3 koodu n'okpuru ahịrị 93 aimet-model-zoo/aimet_zoo_torch/quicksrnet/evaluators/quicksrnet_quanteval.py
2.5. Isi ihe dị na ọrụ load_model
ụdị = load_model (MODEL_PATH_INT8,
MODEL_NAME,
MODEL_ARGS. nweta (MODEL_NAME). nweta (MODEL_CONFIG),
use_quant_sim_model=Eziokwu,
encoding_path=ENCODING_PATH,
quantsim_config_path=CONFIG_PATH,
calibration_data=IMAGES_LR,
use_cuda=Eziokwu,
before_quantization=Eziokwu,
convert_to_dcr=Eziokwu)
MODEL_PATH_INT8 = aimet_zoo_torch/quicksrnet/model/weights/quicksrnet_small_2x_w8a8/pre_opt_weights
MODEL_NAME = QuickSRNetSmall
MODEL_ARGS. nweta (MODEL_NAME) . nweta (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
Biko dochie mgbanwe ndị ahụ maka nha QuickSRNet dị iche iche
2.6 Ngbanwe nha ụdị
- "Input_shape" na aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/model_cards/*.json
- Ọrụ ime load_model(...) na aimet-model-zoo/aimet_zoo_torch/quicksrnet/model/inference.py
- Oke n'ime ọrụ export_to_onnx(..., input_height, input_width) site na "Bupu gaa na ONNX nzọụkwụ REVISED.docx"
2.7 Tinyegharịa 1.6 ọzọ maka mbupụ ụdị ONNX
3. Tụgharịa na SNPE
3.1. Tugharia
${SNPE_ROOT}/bin/x86_64-linux-clang/snpe-onnx-to-dlc \
–input_network model.onnx \
- quantization_overrides ./model.encodings
3.2. (Nhọrọ) Wepụta naanị ọnụọgụ DLC
(nhọrọ) snpe-dlc-quant –input_dlc model.dlc –float_fallback –override_params
3.3. (IHE dị mkpa) ONNX I/O bụ n'usoro nke NCHW; DLC agbanwere n'usoro NHWC
Akwụkwọ / akụrụngwa
 |
Akwụkwọ ngwa ngwa arụmọrụ Qualcomm Aimet [pdf] Ntuziaka quicksrnet_small_2x_w8a8, quicksrnet_small_4x_w8a8, quicksrnet_medium_2x_w8a8, quicksrnet_medium_4x_w8a8, Aimet Nrụmọrụ Ngwa Documentation, Nrụmọrụ Tool Documentation, Toolkit. |
