dmlc/gluon-cv

How to use a trained params to predict?

Open

#229 opened on Aug 2, 2018

View on GitHub
 (9 comments) (0 reactions) (0 assignees)Python (5,610 stars) (1,224 forks)batch import
good first issuequestion

Description

Hi. I am a new one here. It is a awesome lib for CV, thanks for your great work. Now, I am encountering a problem that I trained ssd_300_vgg16_atrous_voc on my own dataset, the dataset has 4 classes. After training, I got a "ssd_300_vgg16_atrous_voc_best.params" file. When I want to load the parmeters and predict on new images, it return an error, logs as it, AssertionError: Failed loading Parameter 'ssd0_convpredictor4_conv0_weight' from saved params: shape incompatible expected (126, 0, 3, 3) vs saved (30, 512, 3, 3) The command I used as follow, from gluoncv import model_zoo, data, utils from matplotlib import pyplot as plt import mxnet as mx ctx = [mx.cpu()] net = model_zoo.get_model('ssd_300_vgg16_atrous_voc', pretrained=False) net.load_parameters('ssd_300_vgg16_atrous_voc_best.params', ctx = ctx)

Thanks for any help.

Contributor guide

Tech stack
python
Domain
machine learning
Issue type
documentation
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
under 1 hour
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
PythonMXNet basics
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60
Research direction
Investigate how to properly load a fine tuned model with a custom number of classes. The error indicates shape mismatch due to different number of output classes. Check the documentation on loading partial weights or using net.initialize() with num classes parameter. Look at comments in issue #229 for suggested solutions.