data2/text/DLP/1016368.txt

Ticket Name: TDA2HF: About import tool sampleInData

Query Text:
Part Number: TDA2HF Hello， I have two questions about import configuration parameters: (1).does sampleInData support video input or multiple images ? (2).if sampleInData support,how to set and how to prepare the video and images? and other question about caffe-jacinto quantize test: I test the imagenet_jacintonet11v2_iter_160000.caffemodel and set quantize: true in deploy.prototxt, save the output of pool5, and then remove quantize: true only,save the output of pool5 again, the results are same."quantize: true" doesn't seem to work.So if quantize: true does work,What happens to the output? Thanks, chen poca

Responses:
Hi Chen poca, 1. You can set "numSampleInData" variable in the import config file and set "numFrames" variable in the infer config file for running multiple images. 2. You can concatenate multiple images to prepare the multiple image input The test results with and without quantize are available here in "caffe_jacinto_models\trained\image_classification\imagenet_jacintonet11v2" folder in the github (https://github.com/tidsp/caffe-jacinto-models), please refer to "run.log" in "test_quantize" and "test" folders. Thanks, Praveen

hello, thank you for your replying. my reason to setting numSampleInData variable >1 is for import process not for inference.if numSampleInData>1,can I set sampleInData to multiple images?

Yes

so how to set multiple image to the sample InData ？using the txt file to list image path？or concatenate multiple images to one .y file？

Concatenate multiple images to one .y file

Sorry for the late reply and thank you for your answer.I have another question,if using multiple images to the sampleInData,would this improve the effect and generalization ability of the quantization?

No