TypeError: Cannot read property 'length' of null (dilations is null in depthwiseConv2dNativeBackpropInput_)

System information

• From the simple-object-detection example
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): tried both on Mac OS and Windows and Windows WSL
• TensorFlow.js installed from (npm or script link): npm library
• TensorFlow.js version (use command below): tested on ^2.6.0 and ^3.3.0

Describe the current behavior Playing with the simple-object-detection example. Ran npm run train and the following happens

Step 1: (Phase 1 of 2: initial transfer learning) - this runs without any problems

  const tBegin = tf.util.now();
  console.log(`Generating ${args.numExamples} training data...`);
  const { images, targets } = await trainingData();
  const { model, fineTuningLayers } = await buildObjectDetectionModel();
  model.compile({ loss: customLossFunction, optimizer: tf.train.rmsprop(5e-3) });
  model.summary();

  // Initial phase of transfer learning.
  // console.log('Images Shape:', images.shape);
  // console.log('Targets Shape', targets.shape);
  console.log('Phase 1 of 2: initial transfer learning');
  await model.fit(images, targets, {
    epochs: args.initialTransferEpochs,
    batchSize: args.batchSize,
    validationSplit: args.validationSplit,
    callbacks: args.logDir == null ? null : tfn.node.tensorBoard(args.logDir, {
      updateFreq: args.logUpdateFreq
    })
  });

Step 2: (Phase 2 of 2: fine-tuning phase) - this is where it breaks

  // Fine-tuning phase of transfer learning.
  // Unfreeze layers for fine-tuning.
  for (const layer of fineTuningLayers) {
    console.log(fineTuningLayers);
    layer.trainable = true;
  }
  model.compile({ loss: customLossFunction, optimizer: tf.train.rmsprop(2e-3) });
  model.summary();

  // Do fine-tuning.
  // The batch size is reduced to avoid CPU/GPU OOM. This has
  // to do with the unfreezing of the fine-tuning layers above,
  // which leads to higher memory consumption during backpropagation.
  console.log('Phase 2 of 2: fine-tuning phase. Top Layer Name:', topLayerName, 'Group Names:', topLayerGroupNames);
  await model.fit(images, targets, {
    epochs: args.fineTuningEpochs,
    batchSize: Math.round(args.batchSize / 2),
    validationSplit: args.validationSplit,
    callbacks: args.logDir == null ? null : tfn.node.tensorBoard(args.logDir, {
      updateFreq: args.logUpdateFreq
    })
  });

I get this error after model.fit(images, targets, {... is executed:

I debugged down a bit further - turns out that somewhere at the kernel input (I think??), it’s giving a dilation value of null (should be number or number[]).

Please advise, I’m going off the tfjs-examples repo simple-object-dectection https://github.com/tensorflow/tfjs-examples/blob/master/simple-object-detection/package.json

…then just running npm install and npm run train

[opiepj]

Still broken, but found a workaround (though probably the model is not even transfer learning)

1. train.js change the tensorflow import from @tensorflow/tfjs to @tensorflow/tfjs-node
2. replaced const topLayerGroupNames = ['conv_pw_9', 'conv_pw_10', 'conv_pw_11']; to const topLayerGroupNames = [ 'conv_pw_11'];

Both changes are required, one without the other leads to the same problem. I have no idea why this fixes it.

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No